Professional Documents
Culture Documents
435Practical Reservoir Simulation
on .
AP API
ov
¥ 06
3
t .
Eos
5 24° APL
z
2
Fou
o2
oo
°
e 2 08 ae cry
Observed Oll Volume,
Fig. 19-15 Coking Corrections
The volume and weight ofthe total sample is meas-
ured first, Ths can be done by a variety of methods. Tae
oil and water are extracted by using toluene as a solvent
as well as heat. This is done with an electric heater, for
safety reasons. The water is condensed and measured in
a graduated container, Note that this measurement is,
made at a specific temperature. At high temperatures,
water is driven olf from clays and other minerals. Since
the specific gravity of water and the bitumen are known,
the volume of oil ean be calculated by subtraction
For additional mones the volume of
oil can also be measured. Common
sense indicates that this should result in a
carey mess lef in the sample and that the
solume of oil recovered from the con
denser will be less than the volume of
bitumen that was originally in the sam.
ple. Tiss, infact, the ease. The volume
Of oil is adjusted Sy cooking known vol-
umes of the bitumen and toluene and
determining a correction factor. An
example of such correction factors is
shown in Figure 19-15.'2
Some of the descriptions indicate that
the total volume of the sample is deter
cnined by immersion in mercury, and this
will compensate for the gasfilled portion
of samples. Normally mercury immersion
destroys samples. Some references seen
by the author say this does not present a
problem and the mercury can be recov:
ered in the extraction process. Mercury
vapor is highly poisonous—2t least
according to the spy novels, The labs vis:
ited by the author usualy have watery
col mixtures on top of the mercury cok
lection trays to stop vaporization
Determining correct saturations
from logs and calibrating to core
Woodhouse describes the correct
method of calibrating logs and core." As
Ys acommon theme in this book, much of
the problems with reservoir simulation
have litle do with the actual simulator
‘The problem, more often than not, is get
ting the correct input data
The basic assumption is simple.
Relatively ite bitumen is displaced from the core due to
its high viscosity. The author questions whether this is
completely true;some litle bubble of evolved gas is going
to cause a spur of bitumen, The oil actually shrinks with
gas exsolution, soit will have shrunk about 1-2%. In prac:
tical terms, these are tolerable errors. The log calcula-
tions then are tuned to weight percent bitumen,The ertors in core attalysis saturations
io not affect the correct calculation of
original vil in place. The increased
porosity compensates for the filtrate
invasion in cores. This does become sig
nificant, however, in thermal recovery
caleulations—ie,,simulations—where
the extra connate water must be heated
as part of the process. Therefore, it is
important wo get the actual porosities
and water saturations correct.
One ther significant item should
be mentioned. AS shown in Figure
19-16, uncemented formations typical
ly have lower values of m, the cementa:
tion exponent in the Archie equation."*
Itis also known that n. the saturation
exponent, normally follows m, This can
be confirmed in a number of ways. A
log analyst can use a Pickett plot on a
waterbearing zone to empirically
determine i. In addition, itis possible
to have the sands tested under overbur-
den conditions.
The following are some actual field
results determined by experiment on
cores obrained from the Hangingstone
area. Examples of this data are shown
in Figure 19-17 and 19-18, The porosity
and saturation exponents from the com-
plete series of tests were tabulated in
Table 19-1
The tests were conducted at overbur
den pressure, Since these cores do not
fully recover from dilations, the: lower
water zones from the adjoining oil sands
leases were analyzed by ARE using a
Pickett plot. An in situ value of 1.35 was
determined form for this area, This value
is consistent with the lower range of the
data in Table 19-1, The following caleu-
lation was based directly on the core
data, Note that t cannot be determined
from a Pickett plot
Reservoir Simy
Therma
ps
Porosity 0%
Fig 19-16 Generalized Variation of Formation Resistivity Factor
with Cementation
Table 19-1 Cementation (m) and Saturation (n) Exponent
Values from Various Athabasca McMurray Formation Sand
Special Core Tests
Welt m n
5-13-84-11W4M 1.45 181
4-15-84-12W4M 147 1.70
15-.20-84-11W4M 1.35 181
1429-84-11W4M 1.36 1.70
13:27-84-11W4M 1.38 1.83
Average 1.40 itPractical Reservoir Simulation
Fig. 19-17 Formation Resistivity actor em —Actual Lab Dara
Sp = axRw/g"xRe 92a)
52 = (10x0853)/
((0.30)2 x 105.31) (19.2)
= 300%
= (10x 0853) /
(0.30)! 105.31) cise)
Se = ITN%
Therefore if. were incorrectly assumed, then
the S,, would have been 30% instead of 17.1%, Using the
revised values of m and n will significanty reduce the cal
culated in sit water saturations
Air permeabilities
The standard design process for SACD was devel
oped on the UTF projects. The predictions of perform-
ance were based on core permeabilities derived direct-
ly from air permeabilities on dilated cores. From the
viewpoint of a conventional reservoir and simulation
198 <
Fig, 19-18 Formation Resistivity Factor n—Actual Lab Data
engineer, this approach is wrong on the face of it
Interestingly one of the statements Irom an earlier paper
(SPE 21529) is most telling:
Prior to Phase A operations, in-house
opinions of what the bulk effective venticat
permeability ofthe formation woul prove to
be ranged over wo orders of magnitude,
resulting in correspondingly varied estimates
of economic viability. The central abjective
of Phase A test was to resolve this uncertain-
1y by direct measurement of the SAGD rate in
@ minimal volume of reservoir'®
Given these comments, itis indeed fortunate that the
pilot went forward, The substantive contingent of geot
echnical engineers and the extensive geomechanical
instrumentation suggests that the use of higher than
Lndisturbed liquid permeabilities was not accidental,
Note that substantial experience in the area of cyclic
steam injection also supports allowance for the effects of
dilation on permeabilityLiquid permeabilities soe
There is experimental data available
that shows the difference between liquid
and air peemeabilties. Again, this data is
trom Hangingstone, This suite of data is
unusually complete. i includes. perme-
abiliy or cleaned and uncleanesamples
and both air and liquid permeability data.
The liquid peemeabiliies were deter —
mined at net overburden pressure (NOB)
A simulated overourden pressure of 5000
kPa was used
‘Tee are two factors 19 consider in
evaluating the eects of NOB >
Normally. horizontally dled
plug is placed vertically ina taxi
al apparatus The fective pressure
applied surrounds the entire sam-
ple, eved in what would be the ver-
tical sides ofthe core plug AS a result, NOB tests
‘ypically provide stresses that are haher than that
which woud actualy exist in the ground
Consequently, NOB tests tend to report lower per
sveabily than they should
Most unconsolidated samples do not behave elas
tical Hence, aot al ofthe unloading effects are
replicated by reloading the sample Typically the
grains do not return to as compacta form as they
would have been in sit
To some degree. these swo factors may cancel out
however, it is not easy to estimate which factor domi-
nates. Thus, some care (educated guess) must be used in
interpreting these results
The experimental results are shown on Figure 19-19.
In total, 25 plugs were tested for liquid permeability The
associated air permeabilities are alsa shown on the plot.
The center ofthe grouping of the dark diamonds is locat
ed in the same position as seen on previous plots of air
core permeability versus core porosity. The light squares
cleatly show the difference in permeability measured
with liquid and with overburden pressure.
The ratio Of kyy/iggy has been plotted on Figure
19-20, As can be seen from the histogram, the average
correction factor willbe in the range of 0.25 t0 0.30, The
arithmetic average was calculated to be 0.249, This is a
Useful indicator of how much conection should be
applied to core air permeabilities to derive in situ reser
voir permeabilities.
satno8 oh,
iquid er
core Porosity traction
Fig, 19-19 Net Qverburden (NOB) Liquid Permeabalt
1980 Faction oF Air Permeability
Documented history matches (UTF Phase A and 8)
did not utilize such corrections. It may be concluded that
there must be some significant mechanism occurring ia
the reservoir that causes air permeabilities to be repre
sentative in SAGD modeling.”
Geomechanical effects
‘The justification for the beneficial effects of geome
chanics is based on:
extensive documentation in the Hiterature
laboratory results
‘numerical simulations
field observations
This combination must lead to the conclusion that oil
sands reservoirs are being affected by geomechanical
effects. A background explanation will be made of gec-
mechanics in the following.
Geomechanical literature
Geomechanics jg not an area that is currently includ-
‘ed in standard petroleum engineering programs. This isa
classic area witere crossdisciplinary understanding is
critical, much like the link between reservoir simulation
and geology. Some university programs have combined
these two disciplines, The following list follows up on ear
lier references to literature on the subject and summa:
rizes a portion of what is available. This material conces
trates on simulations.440 <
cal Reservoir Simulation
Espinoza and Mirabel (1988) used
geomectnics to simulate com
paction and subsidence associat
ed with heavy oil reserirs in
Venezuela, There is considerable
expertise within POVSA with
regard to geomechanics
Untortunatel only a small
par of the Spanish iterature
is translated."
Abou Kassem and Faroug Al
(1985) recognized the need fora.
coupled reservoirigeomechanics
model." Later Tortike and Faroug
Ali (1991) described their attempts
to predict numeriéally the failure
of oil sands due to steaminduced
stresses. They constructed a 3D
stearnwateroilgas model incorpo-
rating geomechanies Plastic
strains and volumetric strains were
found to increase the permeability
and porosity ofthe materia
Tonite, Faroug Ali etal (1994)
used their model to simulate
Number Of Occurenes
cyclic steam stimulation? °
‘A model for oil sands production
was developed by Wan et al
(1989) to describe the effective
stress train behavior of oil sands.
Although the model was devel
oped for casing/formation interaction studies,
ithad applicability to other geomechanics
problems since it incorporated thermal effects
and strain softening with dilation at low
confining stresses.*
Yaziri (1985) developed another model, which
could be calibrated with existing geomechanical
laboratory tests."
Beattie et al, (1991) and Denbina et al. (1991)
utilized a porosity dilation model coupled to a
permeability enhancement algorithm to implic-
itly model prescribed deformation regions
(Gimulating dynamic horizontal fractures) in
cyclic steam stimulation (CSS) history matches
of Ess05 Cold Lake wells. These firsttime suc-
cesses at Cold Lake history matching were
largely attributed to the approach of using a
thermal reservoir simulator implicitly coupled
Petro-Canada Hangingstone Special Core
Overburden Kliquid v. Kair
s
9
s
ot se ° ° & s oe
Kh liquid, NOB J Kh air
Fig, 19-20 Ratio of 4 yoy/Kl ge Hangingstone Samples
toa deformational model (relative permeabill
ty hysteresis was also employed). This follows
from the mechanistic observation from these
studies that formation compaction is by far the
dominant mechanism supplying drive energy
to the early cycles of the CSS process; solution
gas drive was the most important of the
remaining mechanisms. Significantly, gravity
drainage accounted for litte of the oil pro-
duced in the first two cycles, but increased in
importance subsequently?
Chalaturnyk and Scott (1991) modeled the geo-
mechanics of Phase A of the AOSTRA UTF pilot
project. Thermal values from a reservoir simula
tor were fed into a geomechanical model,
‘which used the temperatures to calculate an
equivalent single-phase hydraulic conductivity.
‘They found that, for multiwell patterns, zones of
scant staan inlfailure within the cold reservoir might occur
between the well pairs, rather than near an
injector or producer well
Settai et al. (1992) examined the geomechanical
aspects of fluid injection into oil sands for the
PCES cyclic steam stimulation project. Dilation in
the fracture failure region was identified as one of
the critical design parameters The result was a
‘combined laboratory and simuiation study to
determine the compressibitiy and dilation values
for numerical simulatic
Singhal et al. (1998) discussed the effect of forma-
tion dilation during steam injection if some
threshold pressure was exceeded in the forma-
tion During production, thedetormation was not
entirely reversed. Etfective permeabilities were
increased, enabling largerthan-expected injectivi-
ty and productivity:”
Satriana et al. (1998) modeted an Indonesian
steamflood and demonstrated that oil recovery
can be accelerated due to increased injectvity
because of dilation.A coupled reservoir/geome-
chanics simulator was used. Results showed per-
meability enhancement caused by reduced effec
tive stress forall injection cycles. Field measure
ments from tiltmeters were used to back calculate
increases in reservoir porosity.and these had
increased from 33% t0 a porosity of 38%.A history
match showed the existence of permeability
enhancement due to the reduced effective stress
and limited fracturing in later injection cycles.
Yuan et al, (1999) developed a model tor worm:
hole propagation in cold production from heavy
oil reservoirs. These seiteroding cavity channels
in poorly consolidated sanidstones are responsi-
ble for the unusually high productivity of wells,
Daily production rates are generally proportional
to the cumulative sand production, indicating the
‘wormholes continuously propagate farther into
the reservoir’?
Dusseault et al. (1994) addressed the geome-
chanics of the wormhole phenomenon, as well
as the growth of a remolded zone of dilated
material around the wellbore. Both resulted in
sand production. They recognized the coupling
of sand geomechanics with fluid flow models
as being at the forefront of engineering in the
oil industry."
Taerma!l Aeservoir Simulation
Shen, (2000) of imperial Oil conducted a
suinercal investigation of SAGD process using a
single horizontal wel, He included a dilated zone
around a horizontal well witha rokicedt eapllary
pressure (more on this isue late) att enhanced
permeability A parametric sensitity was done on
diferent geomechanical properties Oil rates were
found to be five times greater than forthe base
case without any dilation.
Ito, Ichikawa.and Hirata (2000) examined the
SAGD steam chamber growth atthe UTE Phase 8
‘and Hanigingstone Phase [ projects:
ito et al. di further work at Hangingstone, such as
resolving anomalous thermocouple readings by
including a high-permeabilty zone laterally from
the injection well JACOS has used high steam
injection pressures of 3,000 kPa withthe intention
of causing the ol sands to fail in shear Steam
chamber rise rates were significanty higher at
Hangingstone than at the UTEEaty growth was
reported 1 be 1,0 mfday as compared to 0.1
infday atthe UTEThe authors interpreted this to
be due to geomechanical effects. to et al. (2002)
concluded that elevated operating pressures. as
high asthe facture pressure. appear to be
required to achieve feasible growth of the SAGD
‘steam chamber.
AicLellan et al. (2000) used geomechanics to
study the caprock integrity of the SAGD process,
rather than the SAGO process itself. Here,a cou-
pled reservoirgeomechanical simulation was
done to examine breakthrough by steam theough
a caprock intersected by a weak discontnuity*
Wong and Li (2000) developed a stressdlependent
model for permeability changes in oil sands due
te shear dilation. Their model dfs from other
approaches in that they predict the permeability
‘changes in one direction under continuous
shearing, The results an anisotropic perm-
eability change.
Settari, Walters, and Behie (2000) described their
geomechanical reservoir modeling of diverse
problems such as injection into oil sands,com-
paction problems inthe North Sea,and brine dis
posal They concluded that geomechanical mod-
eling was feasible on a full field scate.*
Denbina et al. (2001) concluded that geomechan-
ical effects were essential in matching foamy oil
Dehaviorusing permeability enhancements
oarPractical Reservoir Simulation
(made functionally dependent on local pressure
drop in the simulator) #0 provide a realistic
method to model wormhole growth. They also
found that suppressing gas relative permeability
to simulate gas entrainment—was essential in
abating pressure decline and maintaining a high
overall system compressibility.”
Geomechanical behavior
Oils sands ate a frictional material; ie. they derive
their mechanical strength from the frictional resistance
ofthe sand grains. Because they are uncemented by cal
careous of siliceous adhesions at sand grain contact
points, they are extremely weak when unsupported by a
confining stress. The bitumen, being a highly viscous
fluid, is unable to provide any strength other than under
rapid loading,
Coffee test (confining stress)
ARE uses an effective demonstration during presenta-
tions. A vacuum packed package of ground. coffee is
used, The grounds are a granular material and the vacu
‘um packing provides a confining stress of atmospheric
pressure, Vacuum packed coffee is generally packaged in
a form similar to a brick. Most people can stand on one
of these bricks. There will be some deformation with a
person standing on it;however, if you take a pair of
scissors, ora knife, and puncture the package, the
“brick” will immediately collapse into a loose pile
of ground coffee that cannot support any weight
‘without deforming completely. This isthe effect of
confining stress.
Effective stress
Since the oil sands are a frictional material, its
strength is highly dependent upon the effective
confining stress. Effective stress is the portion of
the total stress in excess of the fluid pressure.
o=0-p (193)
where
o' =effective stress
c= total stress
luid pressure
a2 <
The general concept is shown in Figure 19-21, Stress
isa vector quantity therefore for the three principal stress
€5, 6, 6,6, there are three effective stesses, oo! 6,
Strength and stiffness are functions of the effective con.
Fining stress, o', As the effective confining stress increas-
es, the strength and stiffness also increase,
Soil mechanics
The content of these comments is related directly to
the response of oil sands to loading and relies on con:
cepts developed in soil mechanics. Since this Feld of
study is unique to civil and geological engineering, a
brief description has been provided.
interlocked structure
‘The fundamental control of petmeability is the struc
ture of the sand. In ¢ypical biturnen deposits, the sands
are fine grained and the grains af sand are deposited
with the long axes of the grains lying horizontal. In
ny cases, the structure of the sand has been the
result of deep burial in the past, rather than glaciation,
and results in compaction and some diagenesis. The net
result is that the sand structure consists of a matrix of
interlocked grains.
ig, 19-21 Concept of Etective Sess
ans asta a lkWhen subject to loading, the possibie intergranular
behavior includes:
grain overice
grain shearing
‘grain rotation
Srain compaction/expansion
Because of this granular structure, these noncement
‘ed (unconsolidated) sands do not have any substantive
tensile strength, Further, determining the bulk properties
‘of granular material requites that the original structure of
the grains not be disturbed. This requires that samples be
carefully obtained and handled,
‘To date, nv reliable methodology has been developed
to reproduce the same sand grain structure once it has
been disturbed.
Material testing
Methods were developed in civil engineering to quan-
tify the strength of unconsolidated samples. The 10
methods developed were the direct shear apparatus and
the triaxial cell. Basic depictions of these two apparatus
fs are shown in Figures 19-22 and 19-23."
Drained and undrained tests
‘These tests can be run in two fundamentaly diferent
ways. in the First method, pore fluids are allowed (©
escape. which is termed a drained test. in the second
method, the pore fuids are not aliowed to dissipate. This
istermed an undraiged test. This has a profound effect on
the bebavior of the tested materials In an undrained test,
ore pressures will initially support the majority of the
incremental load, In a drained test, sufficient time is
allowed forthe pore pressures o dissipate during loading
The speed of drainage is a function of permeability
Typical responses
The results from these types of tests are shown in
Figure 19-24 on the upper lett side. The testis controlled
by strain, and the Joads are read from a strain cell.
Otherwise, samples will explode if controlled ty load
when the material softens, Note that # represents stain,
which is horizontal for the fitst apparatus and vertical
(axial) for the second apparatus. In a similar fashion, the
Ab represents either vertical dilation or circumferential
dilation for the two different apparatuses respectively. In
the generalized figures, a clear distinction is made
between loose and dense sands. The dense sands show a
peak behavior, ater which their load-bearing capacity is
hermat Reservoir Simulation
reduced. Lanse sands do not show a loss in peok steer.
Before complete shear failure can take place, the inter
locking of sand grains incense sand must be overcome it
addition to the frictional resistance atthe point of contact
After a peak stress is reached, at a low value of shear dis
placement, the degree of interlocking decreases. and the
shear stress necessary to conttinue shear displacement is
reduced. This is termed strain softening
Dilation
The lower diagram un the left sie shows the changes
in volume, which result from shearing and dilation. The
structure of dense sands expands, which has the effect of
increasing porosity and perteability. The loose sands
contact with strain, which has the effect of lowering
porosity and permeability.
Failure envelope
Muitiple tests can be combined to determine a failure
envelope, as show on the right side ofthe diagram. Tas
is termed a MohrCoulomb failure envelope. This materi
al is also presented in Figure 19-25.
Alternate representation of failure envelope
Mohr’ circle can be used to define stress as a combi-
nation of shear and normal stresses for any plane. By
using the properties of this relationship, itis possible to
represent the site of stress with a single point. The co»:
struction is shown Figure 19-26.
The left oF yaxis is p” and the lower or xaxis is the
average stress. By using this constnuction, a stress cond
tion can be described with a point rather than with the
complete Mohr’ circle. This technique makes it easy to
plot the stress history (path) of samples or points in a
reservoir simulation,
Stress path testing
‘The testing techniques described previously are usu-
ally conducted with a constant confining pressure. Itis
possible to conduct testing under different stress paths
by varying the confining pressure by computer control
In this manner, the stess history of an element in aheavy
oll reservoir can be approximated. This type of triaxial
testing was used in the Touhidi-Baghini thesis (1998).” In
these cases, the vertical stress (9,") divided by the con-
fining stress (a,’) can be plotted against axial strain, The
maximum value of 6," is then considered to be the
point of failure.
vanPractical Reservoir Simulation
Normal force
Porous (or solid) plates
al
Loading plate Drain je
\ LI
: SAY} | ter
fae fice,
Drain
Porous (or solid) plates:
Fig, 19-22 Direct Shear Stress Apparatus
Normal force
(permeatity measurement) ——
A
Pressure port
Porous er solid plate
Rubber Le
membrane 7
ydrosastic
pressure
Z|
fr
Drain or port for |
Pressure measurement) 4
Volume change fav)
measured rom hurd
‘volume expelled
(at constant pressure)
Fig. 19-23 Thaxial Cell Apparatus
atc
Constitutive models of
sand behavior
One of the major problems readily
apparent from the previous desceiption is
how to describe the physical behavior of
these sands. The use of the Mohr Coulomb
model to determine the failure point has
been discussed. In fact, there are a qum.
ber of different ways of describing how
sands change. One model is called the
elasticplastic model, another isthe hyper:
bolic model. These two are the most com-
mon methods used: there are, in fact, a
thumber of others. which will not be dis-
cussed in this text. These methods con
ccentrate on bulk deformation of the sand
and how much load can be carried
However, for reservoir engineering
purposes we are interested in a litle
more. How does one tie the loads and
the changing effective stresses (as a
result of changing reservoir pressures)
together with porosity and permeability?
This is not easy, There are a number of
ways to do this. One will be discussed in
the following
‘A triaxial test apparatus can be modi-
fied to simultaneously measure fluid flow
through the failing specimen. In this man-
net, permeability can be related to stress-
es and axial and volumetric strains.
Obtaining samples that are undisturbed
canbe very difficult
TouhidiBaghini (1998) and Scott
(1997) studied the change in absolute
permeability of oil sands while specimens
were undergoing shearing. They tested a
sample of sonbituminous MeMurray
Formation sands excavated from a natural
river outcrop. Specimens were obtained
parallel and perpendicular to bedding
An example of the permeability increase,
relative to the initial absolute permeabil-
ty is given in Figures 19-27 and 19-28% *
At volumetric strains of 4%, roughly cor
responding to the peak strengths of the
samples, the vertical permeabilities
increased by 100%. Some specimens had
an increase in absolute permeability ofShear Stress vs. Lateral Strain Mohr Coulomb Failure Envelopes Volumetric
t yo
o
Fig, 19-24 Shear Test Results
Failure Envelope |
}<—o;
Fig, [9-25 MohrCoulomb Envelope
40-04) Modified Failure Envelope.
_—
S
3
Fig. [9-26 Allerative Representation of Mohe-Coulomb Fai
Reservoir Simulatio
Strain vs. Lateral Str
‘Stress Point a
Hotrod
o
ure EnvelopePractical Reservoir Simulation
one order of magnitude during the test
Permeability increases were attributed to
dilation and shearing, Dilation resulted in
2 pore volume increase and widened
flow paths, and shearing increased pore
connectivity,
The horizontal/vertical permeability
anisotropy was found to be a factor of 1.4.
For a 2% volumetric strain, the vertical
permeability increased by 40%, and the
horizontal permeability increased by 20%
This inctease in vertical permeability has
extremely significant implications for the
advance of the steam front in SAGD.
Due to the unique sampling process
used, this is probably the most accurate
data available on how permeability varies
with shearing.
‘The previous data considered the case
where the bitumen is heated and may
flow. However, shearing can occur in the
cold bitumen. In this case the bitumen
remains immobile, The change in the vol
ume of sand and porosity still causes
changes in fluid flow properties.
According to the data of Oldakowski,
which is shown in Figure 19-29, the per
meability to water is also known to
increase with shearing" This data differs
from the previous data in ¢hat different
sttess paths were used and these tests
were performed at in situ cold reservoir
conditions. The increase in water perme-
ability is substantive with shearing, This
leads naturally into a discussion of relative
permeability
Relative permeability
‘There is considerable debate inthe fit-
erature regarding relative permeability
curves for bitumen reservoirs. There are a
number of relevant issues
whether the shape of the oil curve
is convex or concave
whether the curves are tempera-
ture sensitive
whether the curves are water or oil
wet, which is reflected in the level
of the water elative permeability
45
MO Experiment Ress.
oxen Carman. 6.9 786
Chaaabeigg = 5.6) = 1475
Con 59 ,
tna
Fig, 19-27 Effect of Shear on absolute Vertical Permeability
Moe experimental Rest Gu
Cchardabels,
Chacaabelas
ek
iit.
————-
nit
‘Fig, 19-28 Effect of Shear on Absolute Horizontal Permeabilityat the residual oil sat
aration end-point as
as the value of 10
he irredue
saturation
whether there a
ple water
teresis eltects associa
ed with changes front
drainage to imbibition
Requirements for
accurate relative
permeability data
Obtaining accurate data
is very difficult and expen
sive, There ate a number of
fundamental problems
The tests must be run
at high temperature Lo
Hong states that the . s
most important area
for performance pre:
diction is the high
pressure gascil rela
Live permeability For accurate measurement of
residual oil saturation to steam, both inlet and
outlet must be maintained at wet steam condi
tions. his is certainly consistent with the observa
tions from the Flangingstone hot watersteam
flood test
For unsteady state tests, maintaining such con-
ditions does not represent a long period and
temperature drit is more easily controlled
However. the procedure for steady state tests
takes considerable time and maintaining con-
sistent conditions is more difficult. Therefore,
most tests are done as unsteady state tests
Displacement (unsteady) tests do not measure
relative permeability directly and require con-
siderable inerpretation
Some ofthese interpretation methods are graphi-
cal, which can be quite sensitive to make correct-
ly and can create substantial error More recently.
there isa trend to use numerical simulation to his
tory match relative permeabilities, Unfortunately it
is necessary to assume the shape of the curves in
this process, which is one of the unknowns. In
fact, this methodology results ir multiple sol
8
fective Permeabiiy to Wake, un?
incopse Unioeding 2 Pore Peer Injection
yo 1S 20S 88
‘Water Saturation, %
Fig. 19-29 Ettect of Shear on Water Permeability —Oldakowski
tions, Steady state data is substantially superior,
but itis expensive. Such tests probably represent
less than onethird of the available data
Capillary presstire end effects can distort relative
permeability curves, There are two solutions to
this. First, use longer core samples, and second,
use high displacement rates. Normaliy.the high
rates required are significantly above reservoir
displacement rates, which can lead to unrepresen:
tative data and to viscous fingering. The later is a
significant issue for heavy oils.
‘Weutability is also an important issue. Cores need
to be cut with bland drilling mud systems. Further,
when dealing with noncemented cores, great care
‘must be taken to ensure that disturbance is mini-
mized and that the loading conditions will be
replicated. This includes overburden loading and
effective stress changes.
‘The chemistry of the oit, particularly the
polarity of the components, can also have a
strong effect on relative permeabilisy Good
quality ive samples—i.e,, with solution gas
entrained—are required.
DunPractical Reservoir Simulation
Itis impossible to solve all of these conditions at once,
although one can design around the majority of them.
Such a program is very expensive and rarely done,
Three-phase relative permeability
‘Ypically. thtee-phase relative permeability has been
calculated using a correlation. The correlation is a variant
of those developed by Stone. Although a standard prac
tice, the correlations have been assembled based on a
limited number of available data sets. None of these
included heavy oil samples. Abou-Kassen and Faroug Ali
make the following comments
Trreephase relative permeabilities are
the Achilles heel of reservoir simulation. This
is more so for thermat simulation. Relative
permeabilities measured in the laboratory
‘are questionable in the light of instability the
‘ory (viscous fingering). Frequenthy the rela:
tive permeabilities become a history match
parameter The laboratory value of end
points are still of value, In thermal simula.
tion, the temperature dependence of relative
permeability should be taken into account.
‘The same can be said of hysteresis. In many
cases, two-phase relative permeability data
4 used in some variation of Stone's model to
oblain threephase relative permeabilities
Myhill has refered to the problem of obtain.
ing the low oil saturations behind the steam
front using Stones model. Once the end
points are obtained, it is easy to use the
Naar WygatHenderson equations. Kaeraie
gives a modified version of these, suggesting
that one way of including temperature
dependence is to make the end points in the
equations temperature dependent. Relative
permeability hysteresis should be included in
the simulation of cyclic steam stimulations.
This can be accomplished in several ways; a
simple approach was discussed by Bang.
Sato has discussed the role of relative per
‘meabilties on thermal simulation."
Again, the low saturations behind the steam front sug-
4gest that the gas- (including steam) bitumen relative per-
-meabilty set is critical to residual saturations,
aaa <
Laboratory scale modeling
Butler and Chow used straightiine pseudo relative
permeability curves to history-match some lab experi
meats. Their conclusions indicated that SAGD simula
tion was insensitive to the choice of relative permeabili
ty curves. They referenced Dake in the development of
these curves
The key assumption in the use of the VE technique is
that the height of the capillary pressure transition zone
is much smaller than the vertical dimension of the grid
block, Although most capillary pressure lab data over.
estimates the thickness of transition zones, most logs
indicate transitions that are at least one meter thick
Since the capillary transition zone is the same size as the
grid blocks (particularly for a lab experiment), it was
inappropriate to use segregated relative permeability
‘curves, Dake outlines the use of pseudo relative perme
ability curve development with finite capillary pressure
transition zones.
Many ofthe water bitumen relations are quite straight
in comparison to conventional light oil relative perme
ability curves. In this case, the history match likely worked
simply because the curves closely resembled what the lab
curves would have looked like—had they been obtained.
Sasaki, Akibayashi, Yazawa, Doan, and Faroug Ali
experimented with different relative permeability curves
to match some of their scaled laboratory models. AS
shown in Figure 19-30, they found that the shape of the
steam chamber was strongly atfected by the choice of rel
ative permeability curves. Their work utilized a number
of different assumptions. They strongly disagreed with
Butler and Chow's conclusions that SAGD is insensitive to
the choice of relative permeability curves. The varying
shape of the chambers would make this point obvious
However, their best match utilized straightline curves.
Again, this may represent rock properties.
Its unfortunate that the latter authors did not sepa-
rately investigate the effects on the gasbitumen and the
waterbitumen curves,
For conventional light oi! modeling, using straight:
line curves will often work, is an accurate representa-
tion of reservoir physics, and can be substantiated by
detailed modeling with multiple layers. These curves,
are rate sensitive and often do not apply in predictions
At higher rates, segregation may not occur to the same
degree, and the capillary pressure transition often is not
accounted for in this process. It may be concluded that
conditions for using VE should be proven before they
are used for predictions.Thermal Reservoir Simulation
(Pas Pee Te ERT HS
[Fob Ea
(a) (0)
Numerical simulation temperature contours (at t= 550 min)
with linear relative permeability functions
{a) wath zero éndpoint saturations, (b) with non-zero endpoint saturations
* Wy
et id
fl “i yf
Temp interval = 7.9°C
Numerical simulation temperature contours(at t = 550 min)
with non-linear relative permeability functions
ig, 9-30 Numerical Simulation Results with DiflerentShaped Relative Permeabiliy Curves
> agPractical Reservoir Simulation
Importance of relative permeability data
Shen investigated the effects of capillary pressute,
undulating horizontal wellbores, and the rates that would
be obiained. In this study, the relative permeability curves
used are as shown in Figure 19-31, This paper also dis
cusses the Importance of countercurrent drainage and
capillary imbibition pressure thresholds. Earlier work by
Edmunds indicated that sufficiently high capillary imbi-
bition pressures could prevent the SAGD process
Earlier work by Adegbesan utilized straightline relative
permeabilities,
kKisman and Yeung modeled the Burnt Lake Oil Sands
Lease. In this case, there was no experimental relative
permeability or wettability data, They indicated that the
UTF simulation was not particularly sensitive to relative
permeability curves, However, they did conduct sensitiv.
ites to relative permeability curve entt-points. The data
they used is shown in Figure 19-32." They perlormed
some different sensitivities with different shaped curves.
The extent of all of the modifications is not clear
However the use of Stone's relative permeability correla-
tions was predicated on the k,, at the irreducible water
saturation being equal te 1, Othenwise, unpredictable va:
10
& &
Relative Permeability. k
02
tues may occur Overall, the effects of their various runs
indicate changes from worst to best case of only
However, they do not seem to have changed the shape of
the curves greatly
It should be noted that Kisman and Yeung used a very
low gasbitumen residuat oil saturation
In the finad analysis, it may be that geomechanics,
numerical dispersion, and heterogeneity have more
effect than the shape of the relative permeability curves,
in ARES experience, one thing stands out—the water rel
ative permeability end-point is increased significantly. In
conventional reservoir simulation, this is usually the
effect of layering. Pseuco relative permeability curves are
easy to generate for history matching. Unfortunately they
are not transferable from one reservoir simulation to
another and are of limited to no use for predictions. tis,
always better to model the heterogeneities directly in the
model with quantitative reservoir characterization,
Ik should also be pointed out that, in his textbook,
Butler notes that he has observed fingering at the top of,
steam chambers, and this displacement is very likely to
WateriLiquid Saturation 5.1
Fig, 19-31 Relatwe Permeability Data Used by Shen
450 <Thermal Reservoir Simulation
tive Permeability
02 04 06 08 10
Water Saturation
04 08 08 10
Water Saturation
Fig. 19-32 Relative Permeability Curves Used by Kisman and Yeung
> 45!ctical Resesvoir Simulation
be unstable. This would be reflected in relative perme:
ability curves that are history matched. The sides of the
chamber appear to be inherently stable,
Relative permeability data available
There is a range of data outlined in a number of text-
books on classic heavy oil reservoir engineering
452 <
i
IFP shows a nutaber of curves.all of which are
waterwet samples, Most of these are for
cemented sanrstones. These curves are shown,
in Figure 19-33."
Exxon (Boberg) showed some curves used as
demonstration examples and analyzes the sensi-
tivity to different curves. ese are shown in
Figure 19M. Exxon’s data also includes gascoil
relative permeability curves!
Gravdrain (Butler's textbook) showed curves for
‘cemented samples. as shown in Figure 19-3555
Butler also showed some changes in relative per
meabilty related to temperature. Note that not all
fein merge
researchers agree on this point It is quite possible
that some samples are temperature sensitive
while others are not
Maini and OXazawa showed relative permeability
cures for silica sand that was packed into 3 core
holder These are shown in Figure 1936.1 this
case, they inferred temperature sensitvit:
However these were unsteady state tess. The enc
point saturations were also forced 10 a common
salue, which may not be cue. They discussed
many of the limitations of history-matehed rela
tive permeability curves.
Pulikar,Puttagunta, DeCastro.and Faroud Ali also
investigated the shape of relative permeability
cures in"Relatve Permeability Curve for
Bitumen and Water in Oil Sand Systems” Their
research used the steady state methodology and
‘a noncemented sample. They found convex oil
relative permeability curves as shown in Figure
19487. Again, note the high ,, forthe wateroil
curve without any steam present. They compared
saa
vote wares
Fig. 19-33 (a, ) FP Heavy Oil Relate Permeabilcy DaraEffect of Temperature on Absolute Permeability
lity at 20°C
| © [alumina 027s | 446
O Fontainebleau Sandstone} 0.240 | 142
© Alumina 00% | 408
|G | alumina | oor | as
z
2
2 | A | Fontainebleau Sandstone | 0.00
z Font pene 0000
é 25075 ~100~«3S~—~«SD~«S
(c) Temperature,°C
Effect of the Effective Stress on Porosity and Permeability
at Two Different Temperatures
T2212
_
T2217
Porosity 0, %
Be
Permeability, um?
=
T=93.8°C Ta938°C
SS eoe
6 10 14 2 4 10 14
(d) Effective Stress, MPa (e) Effective Stress, MPa
Unpublished data, Inet Frang du Patol,
Fig 19.33 (¢-<) Heavy Oil Relative Permeability Data (Subsitiary Etects)
eservair SimulationPractical Reservoir
10.0
001
ee 7 oF
baa OF
Relative Permeab|
or a ae er.)
Water Saturation, % PV
Figure 49. Relative Permezbility
Curves for Becea Sandstone Core
(from Lo and Mungan 1973)
Fig, 19-35 Relative Permeability Used by Butler
Oil Recovery, PV
Steam Flood
Hot Water Flood
0 o¢ 08 12 16 20
Liquids Produced, PV
RELATIVE PERMEABILITY
"00 02 04 06 =e 40
WATER SATURATION
Fig 19-36 Ternperanre-Dependent Relative Permeability
From Maini and Okazawa© Bitumen ( multi—core | |
(© Water ( multi-core }
a Bitumen | single—cove ) |
& Water ( singlecore }
— Best Fit
0.6]
tive Permeability
=
é
02
|
00
ao 02 04 O06 O08 10
Water Saturation
Bitumen-water relative permeability curves
<=. Lloydminster, run 7
‘
Lloydminster, extracted core]
Lloydminster, preserved cove
oa 08 081.0
Water Ssturation
Normalized heavy oil-water relative
Permeability curves trom experiments
‘ig, 19-37 Experimental Curves From Polikar, Puttagunte, DeCastr,
‘and Faroug A
hermal Reservoir Si
the shape derived for a number of
other different experiments This
underscores the degree of vara,
tion found in tests
8. Donnelly used the curves shown in
Figure 19-38 forthe Hilda Lake
SAGD project. His predictions
were outstanding using this data
The initial water saturations were
very high, apparently with no nega
tive effect on oil production. His 5,
on the gastiquid relative perme
ability was very iow,which, as
described previouslyis critical. His
z,, curve drops to zero at aS, of
30%, and his water relative perme-
ability curve extends beyond the
S,-This is somewhat unusual, as
most, Curves would end at S,.
However it would allow for simu
taneous steam and water move-
ments The end of the &,, curve
appears to coincide withthe likely
end-point of the gasliquid relative
permeability curve
Unfortunately, only two sets ofthe pre
vious data included gasditumen relative
permeability curves.
Hangingstone steam flood test
Earlier in the chapter, temperature
dependence of relative permeability end
points was discussed. Figure 19-39 shows
results with these effects from the
Hangingstone area of Alberta. Tis water
flood/steam flood test dates frorn 1984, It
was performed on a fresh state core plug
as well as a companion sample. The com
panion sample was used to derive conver
tional reservoir properties, such as initial
oil and water saturations.
‘The fresh state sample was placed in
a hydrostatic core holder in a temperar
turecontrolied oven, A temperature of
100° Celsius was used with a net overbur-
den pressure of 600 kPa. This would be
equivalent to a depth of approximately
300 meters.servair Simulation
Practical f
Water-oil Relative Permeability
000 010 02% 030 0M0 050 a 070 080 O80 100
(Water Saturation fraction)
Gas-tiquid Relative Permeability
2
3
2
&
Rete Penpenbiny (recto)
& &
10
ono +
000 010 020 «0% 0m 08 «ae 070 08 080 1.00
quid Saturation (rection)
Fig. 19-38 Relative Permeability Curves used by Donnelly for Hilda Lakeermal Reservoir Simulation
Hangingstone Speciai Core
Recovery During Hot Water / Steam Flood Test
300 ~
250
z ts
Q 200 }— 8¢
2 | g2
> 180 | -— ps
5 $e
£ 100 | . - 35
z | isseira ira | Be
50 0.4
0
0 50 100 150 200 250 300
Temperature (C°)
Fig, 19-49 Nangingstone Steamlood Test Results
‘The sample was flooded with hat water at 100" Celsius
and 4,900 kPa to residual oil, and the permeability was
determined. The oven temperature was then raised to 150°
Celsius, the sample was allowed to equilibrate, and was
again flooded with hot water at 4,900 kPa, The pressure
‘was dropped to 450 kPa, and the sample was flooded with
steam. The sample was flooded until oil production
ceased and the permeability to steam was determined.
Alternating water and steam was injected similariy at 200°
Celsius with pressures of 4,900 kPa for steam and 1,500 kPa
for water. This process was repeated once mote at 250”
Celsius with injection occurring at 4,900 kPa for water and
3.900 kPa for steam, The results are outlined as follows:
PS
Péa
2S
120 793
5800mD 34.8%
6,190 mD_ 36.7%
The residual saturations were determined by Dean
Stark extraction. Following this, air permeabilities, porosi
ty, and grain densities were determined. Finally the sam.
ple was resaturated with water and a liquid permeability
determined.
Note that the companion sample saturations sum to
91.3% and not to 100%. With the summation of fluid
method, only water and oil are measured. The likely
source of the difference is air that had been imbibed in
the core during unloading or gas that has been liberated
from the bitumen. The sample on which the flooding was
done was restressed, and this air or gas would most likely
have been compressed out of the core. This makes the
original in situ S, 13.1%, the $, 86.8%, and the S, 25.7%.
Detailed analysis of recovery
itis interesting to analyze closely the recovery associ-
ated with the individual steps in the previous process
Recall that there was an initial displacement with hot
water, followed by three stages at different temperatures
in which hot water was fist injected, followed by steam,
Recovery data is summarized in Table 19-2.
The water sweeps an increment of recovery of 208%
of the OOIP At this temperature, Athabasca bitumen has
a viscosity of more than 200 ep. When the temperature is,
increased, the viscosity drops to about 60 cp, and the
water is able to displace an additional 14.4% of the OOIP
AA this point, flooding with steam atthe same tempera
ture adds another 18.6%. Steam dramatically increases
> 57Table 19-2 Summary of Water Steamflooding Test Results—Hangingstone Athabasca McMurray Sands
Temperature| Phase | Pressure | Iefemental | Cumolative ] Percent | peemeabiity
% Boa | Boe Voune | Bree Vole | 00% [nb
150, steam | 450 0.1480, 538 3140
2) [sean [2900] 0.00 wes
recovery Increasing the temperature and Nlooding with
water adds an additional 11.8% of OOH Switching to
steam adds only a small increment of 4.8%
A number of additional observations can be made, It
seems that higher temperature steam adds to recovery up
to 200° Celsius or 1,500 kPa Conversely, running a steam
chamber below 1,500 kPa will decrease recovery. One
consideration in SAGD design is heat losses. Lower tem
peratures will educe heat losses. However, this evidently
comes al the expense of recovery Temperatures above
200° Celsius and 3.900 kPa do not seem to add to recov.
ery from a core flood point of view
Another interesting observation is that the steam
permeability increases from 321 mD to 465 mD when
flooding is changed from 200 to 250° Celsius. Since
there is no increment in recovery, it would seem that
there should be no change in ol saturation. This is like-
'y an elfective stress effect, The intergranular pressures
are reduced, withthe higher pore pressutes leading (0
lower effective stresses and higher permeability It is
also clear that there is 2 large change in permeability
between the steam and hot water. Note that permeabil-
ily iS independent of viscosity, and, therefore, this
would have to be a relative permeability effect. This
would also indicate that ultimate recovery is more
strongly affected by gas-bitumen relative permeability
than waterbitumen relative permeability.
History match relative permeability curves
Much of the historical design procedure for SAGD
projects was based on the work conducted at the UTE
These curves have become widely used. Originally, the
relative permeability relations used by AOSTRA were
derived on what AOSTRA has described as general expe-
rience. The author questioned a number of the people
directly involved, and it does not appear that experimen:
tal data was available.
‘The history-match curves are, in fact, nothing more
than pseudo relative permeability curves, a topic that
has been discussed extensively in this book. As such
these UTF curves represent a number of different
things including:
‘numerical dispersion
lajering/heterogeneity
viscous fingering
capillary pressure transition zones
production rates
completion configurations
geomechanical effects
all in addition to the fundamental properties of
the rock
Of course, many of these factors are unique to the
UTF project. Therefore, the use of these curves provides
litle assurance that they are applicable elsewhere. This
does not imply that the methodology used was incor
rect. In fact, this was the limit of technology as it exist
ed at the time.
A conference paper by Polikar, Puttagunta, DeCastco
and Faroug Ali summarized a number of relative perme:
ability curves used jn simulation. The curves are shown in
Figure 19-40. Note that there is considerable variation in
the curves. The degree of history matching and the gridblock sizes are not outlined, However, it does serve 10
show the variation in curves used. It seems that some
degree of skepticism is indeed warranted.
= =~ Athabasca
Toole Lake
“Cold Lake
~Lleyemineter
Tanabe 7,
Relative Permeability
of
Water Saturation
a0
ig, 19-40 Normalized Relative Permeability Curves Used in
Simulation—Various Sources
Choosing relative permeability curves
Anumber of dificulties have been outlined previous-
'y As practicing reservoir engineers, a decision must be
made, for better or for worse, Ia unique set of data can-
not be identified, then a range o! sensitivities should be
performed, It should also be noted that, despite many of
these uncertainties, thermal reservoir simulation has
proven to be capable of predicting the performance of
SAGD projects. This is well-documented in the literature
Relative permeability end-point—
connate water
Faced with the determination ofthe connate water sat
uration end-point, ARE summarized a variety of different
data sources as shown in the following:
Hangingstone (an oltset reservoir)
P Data 318%, aug 13%
Hot Water/Steam Flood at Hangingstone 13%
UTE (Offsetting Reserwoir) History Match 15%
Polikar etal. (Technical Paper) Clean Sand 6%
Log Analysis 1%
hermal Reservair Simulati
Maini and Okazawa (Paper) Clean Sand
Bennion et al. Preserved Core
(Ottset) Reservoir
Band 58
Most connate water saturations on cemented rocks
are higher, normally in the range of 25%. Although such
curves have been discussed, they will not represent accu
rately oil sands conditions of connate water saturations.
‘The most accurate data in our view is log analysis pre
pared for the actual lease under study. ARE has rounded
up the logderived S,. slightly to 15%, which is similar to
the majority of existing models.
Relative permeability end-point—
residual oil saturation
Folikar, DeCastro, Puttagunta, and Faroug Ali made an
extensive study of noncemented sand in “Effect of
Temperature on Bitumen-Water End-Point Relative
Fermeabilities and Saturations” They found litle temper:
ature effect. Note that the residual oil saturations in these
tests, which do not include steam, were relatively high—
averaging 46%
Maini and Okazawa calculated all of their end-points
at a residual oil saturation of 20%. Experimental end-
points were not shown in their results
Bennion, Sarioglu, et al. performed tests on stacked,
preserved core material. In general, ARE has found that
careful screening of stacked data in comparison to
solid samples indicates core stacks are not reliable. The
problem is the multiple capillary end effects that exist
at sample junctures. This problem is exacerbated if the
samples do not mesh exactly inside the core holder.
Noncemented cores are less likely to cause this prob-
lem because the material will mold. Their results indi
cate waterbitumen residual saturations of 31.1% and
34.6%, respectively®
Further guidance on this matter is provided by Faroug
Ali/Butler (via previous Butler reference) as shown in
Figure 19-41. Note that the bitumen viscosity of the
Athabasca deposit is about 1.7 million cp at an initial
reservoir temperature of 8-10" Celsius. This falls slightly
off the diagram and to the left. There is no correlation
between original reservoir bitumen viscosity at original
reservoir temperature and residual oil saturation. There is
a range or belt of values across the chart. Note that this is
for a variety of steamfloods. Therefore, this would repre-
sent the effects of various degrees of wateroil and gasoil
relative permeability
onPractical Reservoir Simulation
25
> 204 Aca
z
2089)
x ae S
b esican (Paar ie
pm aoe $y
woh oe
3
«
1o® 101 10? 103 104 108
Oil Viscosity at Ty in ep
Fig 19-41 Empirical Data of Residuat Ou! Saturation
Butler has presented calculations that indicate the
residual oft saturations are lower for the SAGD process
than for conventional steam flooding. He estimates a typ
ical steam chamber to have a residual oil saturation of
0.10 at 2,000 days,
As outlined from the water floadisteam food sample
‘and other data, itis likely thatthe end-point saturation for
gasbitumen and waterbitumen are quite different. There
are other considerations here:
The Stone Il phase relative permeability correla:
tiom was designed with a consistent S,,
In the SAGD process, fluid is tree to drain verticak
'y unlike the lab tests to which frontal advance
calculations have been calibrated. In fact,the
process probably takes place across a thin zone
rather than at a sharp interface,
This displacement zone is probably of the order
‘of one meter or less.This is less than the thickness,
of a single grid block, Therefore, the process will
ot be modeled using the one-meter grid blocks
selected. The relative permeability curves have to
account for this entire zone.
Accordingly, in past simulations, ARE has assigned a
single value for residual oil saturation for both the oi
water and gas-il relative permeabilities.
459 <
Steam-water relative permeability
and capillary pressure
Steamewater relative permeability is more complex
than oilwater permeability in that there can be mass
exchange between the two phases. Experimental investi
gations of this have been done by Home and indicate
that low interfacial tensions and/or mass exchange de
occur 4 case can be made for comparison with ga
condensate relative permeability where liquids drop out
Of the gas, and this alters the formation of retative per
meability curves, This topic was discussed earlier. Thus
fa. there does not appear to have been much research in
this area
‘The results of Horne contradict some earlier heavy
oil steamflood testing and have also been applied or
developed for geothermal research. These issues
seem logical and, like the temperature dependence
of relative permeability, there may not be uniform or
consistent results
Geomechanical effects on mudstones
Geomechanical effects can be expected to occur in
mudstones, As steam condenses on the underside of a
mudstone, the hot water 1S free to penetrate the much
stones. This is not analogous to trapping gas. Gas accu
tmulations are trapped by capillary imbibition pressures
not low permeability Steam would be a nonwetting
phase. In fact, as outlined in Bachy and Undershuitz,
caudstones and shales act as aquitards and are not barr
ers to wetting phase flow—ie., water
Since the water can penetrate mudstones. their per
meability will also increase due to pore pressure effects
as well as shear failure, These effects will be more pro-
nounced where shear forces and dilation are the
strongest, which normally would be near the injector,
producer, and in the area directly above the wel pai.
‘There is also hot plate heating above mudstones,
wich propagates by conduction. This reduces the viscos
ity of the oil above the mudstone and eventually leads to
bypassing ofthe barrier asthe oil becomes mobile.
Probabiy the most convincing evidence ofthis was
demonstrated in the AOSTRA UTF Phase A, where a mud-
stone existed between the injector and the producer
Communication was established and this, in ARES opin-
Jon, represents structural changes to the mudstone fabric
resulting from geomechanical effectsermal Reservoir Simulat
Numerical Difficulties with Thermal Simulation
Numerical diffculies with reservoir simulation for
black oi models are relatively rare. This is aot true, at
least vet, for thermal simulations, The following are trou
ble spots the author has encountered.
Grid-block requirements
Thermal simullations are affected by the choice of grid
size, as are conventional reservoir simulations. However,
since the mectianism is somewhat different, SAGD sim
lations requite much smaller grids than are typical for
other processes. Based on sensitivities conducted by
ARE, an optimal grid size is approximately 1 meter hori-
zontally (j direction) by | meter vertically ( direction).
In the cases described in the following, all simulations
‘were conducted with one SoOmeter grid block along the
horizontal well (direction)
There is @ noticeable amount of error associated with
this grid spacing. The accuracy of output would be on
the order of 10%. Figures 19-42, 19-43, and I-44 show
a grid sensitivity
Nine-point difference schemes
Due to the high mobility of steam and the contrast
with the low mobility bitumen, most thermal reservoir
Simulators include nine-point difference scheines 10 con
vol grid orientation effects, The fatter was discussed in
chapter 2 of this book
The implementations on the simufators used by the
author pertnit nine-poiat differencing in only one plane.
Fortunately the plane can be chosen as either ik or fe
On some ofthe simulations, this seemed to have no
effect an the execution time, One might expect the more
complex difference scheme to increase run time. On the
runs that the author examined carefully, the results
appeared to be essentially the same. Interestingly. the
largest difference seemed to be in improved material ba
ance, which was of significant benefit.
While the runs under discussion can hardly be called
an exhaustive review, it does seem that the nine-point dit
ferencing was beneficial. The results that were obtained
ate shown in Figures 19-45, 19-46, and 19-47. The difer
fences in material balance are shown in Table 19-3
= Steam-Cill Ratio
—— Steam-Oil Ratio
Steam-Oil Ratio (SOR), (injector)/(producer) INS (m?/m?)
20022005
30052008 —2007~—~=«2008~=~«OD
Fig. [9-42 Grid Sensitivity team-Oil Ratio (SOR)
op
8Practical Reservoir Simulation
a
ate So tay)
$62 <
703 me m5 m8 or
‘ime (Oate)
Fig, 19-43 Grid Sensitivity Production Rate and Water Injection Rate
me 08 oe ae 75 27
Time (Oe)
Fig, 9-44 Grid Sensitivity Cumulative Oil Production
ae
8Material balances
Obtaining low material balance errors is much mere
dificult in thermal simulation, Normal black oil simul
tions can usually obtain material balance errors of signit
icantly tes that onesalf a percent and often in hun
dredths of a percent, The thermal simulations that
involved gas and water legs often involved material bal-
ance errors in the 1-2% range. The simulation vendor felt
this was within technical norms, The author was able to
achieve better results with just bitumen and water but no
overlying gas, Thermal simulations generally are not as
stable and have more difficulty in achieving convergence.
‘Thermal shock fronts move convectively, unlike pressure
shocks, which move diffusively Generally speaking, ther-
mal simulations require smaller timesteps than normal
black oil simulations
Thermal simulators also provide an energy balance
ARE has generally found that the energy balance gen-
erally correlates with the material balance—i.e,, what-
ever the level of accuracy reflected in the material bal
ance will correspord to about that ievel in the energy
balance. The energy balance should also be checked
along with the material balance. It also provides some
interesting insight into the overall efficiency of the
SAGD process.
2 SOR econosuy COM (rn
‘Ste
Thermal Reservoir Simulat
‘The majonty of the cost in SAGD is wate! to produc
ing steam. The material balance can be used directly in
the calculation of efficiency. Not all produced heat can
be recovered, however impruved efficiency in recover
ing heat will significantly reduce steam generating costs
Longer run times
Mos SAGD simulations are nun with 2D erosssection:
al models. Considerably more computation is required
with theemai models than with black il models. As out
lined presioust; grd-block sizes have to be quite small
Normally, with conventional reservoir simulations, larger
grid blocks can be used away from the wells. The nature
of the SAGD mechanism does not allow large grid blocks
tobe used, in SAGD, the steam chamber front progresses
by melting a grid block’ oil, thereby precluding the use
of large grid blocks. Some minor changes can be made
by moving from I-to .S:meter grid blocks. Based on the
sensitivities run by the author, thisis not recommended
me
me (Oa)
Fig, [9-45 Nine Point Grid SOR
ionPract.
Mee
Rate SC, (m*/day)
ical Reservoir Simulation
300
= Oil Rate SC balances irf
—— il Rate SC balance2.irf
{ i | | ——- Water Rate SC balance2.irf
—— Water Rate SC balances irf
200
2001 2002 2003 2004 2005 2006 2007 2008
Fig, 19-46 Ninepoint Grid Sensitivity Production Rate and Water Injection Rote
Cumulative 04 SC (9,
05 ma Be on
Time (Oat)
Fig. 19-47 Nine-point Grid Sensitivity Cumulative Oil Production
2008Thermal Reservoir Simulation
Table 19-3 Material Balance Statistics from Grid Differencing (Nine-Point vs. Five-Point Sensitivity)
Without Niae-Point Discretization
Initial Total | Current Total] Net inj/Prod [Net Heat Lost] Error
water [nod | A38E-09 SOE TES DEF
on Guo | var Oe aes SHES SE
cH (amol) | 3508-08 2 ie Cpe OTE
Energy [UY TRIE Se a) Soe | -L06E-T3
With NinePoint Discretization
Initial Total | Current Total [ Net inj/Prod [Net Heat Lost] Error _[ % Error
water [mod | 4386-09 2986-08 “SE *150.07 | _O.ATI6
ou (mon | 4998-08 1386-08 “337-08 TSies06 [OTS
oH (gino [ 538606 SRF Done=00 BSDE-0S | 2156
trey [GD 39S 208614 528bc0 “hae | Sen? | Oar
Steam injection well instabilities
For the injertion well, there are often stability prob-
Jems. Steam has a viscosity more like a gas—in the range
010.015 cp. The viscosity of steam and water are shown in
Figure 19-48. There are very high permeabilities used in
the unconsolidated sands, particularly when accounting
for geomechanical effects. The net effect of this i that
‘when the steam injection rate is controlled, the solution
is very sensitive to the well block pressure, This often
leads to instabilities
oa]
2
Dynamic vacosty In op
2
8
8
[steam | | | |
° 00) 700 300
Temperature degrees Celsive
‘Fig, 19-48 Viscosity of Steam and Water
There are a number of options available, I is possible
to tighten the tolerances on the pressure solution and, in
some cases, the pressute solution for the welt equation
alone. Sometimes this seems to work, and sometimes it
merely increases run times
One of the simulator developers simply changes the
well index to reduce the instabilities. This significantly
decreased run times and improved the material balance,
butt may not be desirable ifthe well will goon a pressure
control later tn his case, the injectivity of the well may be
unrealistically reduced
Having said this, most horizontal sections are not
completely open. Screens are placed across roughly one-
quarter t0 one-half of the horizontal well section
Realistically, conventional reservoir experience indicates
skins are eeal, The author has not found any available
data where the skin has been ineasured.
The author has experienced stability problems on
horizontal wells in conventional oil reservoirs where
there were pressure solution problems on the produc-
ers. In this case, the production was not affected by
‘quite large reductions in the well index. The limiting fac-
tor seemed to be the amount of oil that could converge
‘on that part of the reservoir, It is possible that the sur
rounding formation will control the rate at which steam
can be injected
485Practical Reservoir Simulation
400
Water Rate (ma/day)
8 8
0 2 4 6
400
g
‘a Rae $0 pce
Wiis ils SE ita
3
Time (yt)
Water Rae (maiday)
100
© 2 4
Fig, 19-49 Stability Dificulies with Steam Injection
This problem is very easy to spot on a horizontal
reservoir simulation. A prime example is shown in Figure
19-49, f nothing else, the results are not visually pleasing
[tis probable that a smaller grid block would not solve
this problem, although the author has not tested this
Slightly larger grid blocks might help.
Countercurrent flow
Ina number of runs, the author had difficulties when
there were gas and water over the bitumen. The objective
Of these runs was to match the pressure at the top of the
steam chamber with the pressure in the overlying zones,
‘Thus, the low-density steam was rising while the higher
density water was falling. In addition, there were phase
changes with the water as it was contacted by the steam,
This can lead to numerical problems and very slow
execution times. It appears that the countercurrent flow
caused difficulties for the reservoir simulator, The prob-
em appears to disappear when the fluids enter the
steam chamber, which, if the volume is large, can
‘quench the steam chamber, or when the steam pressure
is high enough to cause displacement of the steam into
the overlying zones.
485 <
Phase shifts
In one of the simulators used by the author, a control
is availabie that limits the number of times the phases can
change from successive iterations. The description for the
use of this control states:
When the reservoir conditions are at sat.
rated values and the reservoir Muids contain
very volatile components, of the ‘otal heat
capacity of a grid block is small, it may hap-
pen that @ very small change in pressure or
temperature will cause phase appearance or
disappearance in a grid block in almost
every iteration during a timestep.
A keyword can be used to limit the number of phase
changes to a specified number This may cause the
numerical performance to improve, but it can come at
the expense of material balance errors,Heat capacities and thermal conductivities
When STARS. a product of CMG. was developed
Sinmuttaneously with SAGD, there was a bug in the calcu-
lation of heat transfer. The simulator identified the grid
block with the highest temperature, a form of upstream,
‘weighting. The transter of heat was then calculated using
the thermal conductivity of this upstream block to the
this led to
adjoining blocks. In some circumstances,
oscillatory behavior and convergence problems when,
contrasting thermal conductivities existed in adjacent
uric blocks.
For this one will find that older data sets,
which are used to develop subsequent simulations, often
have identical thermal capacities and conductivities for
the rock matrix, gas, and water. This step was designed to
avoid instabilities that might occur, The author has found
these common properties in a number of data sets,
Since that time, the calculation has been changed
and now uses a harmonic average thermal conductivity
between grid blocks. In the author's experience, the use
of averaged thermal properties seemed to effect no.
‘ble change in results: hence, itis likely that this
not
practice is no longer required.
Slabbing
Very atten in the early period of the oil production,
there 8 an anomalous spike of oil production
occurs in most runs and seems to coincide with the per:
od immediately after the upward plume of steam his the
top of the reservoir,
‘This appears to representa numerical effect. The oil in
the stack of grid blocks adjacent to the steam plume
seems to melt at once, which leads to the short-term oll
production rate spike shown in Figure 19-50. The author
has termed this slabbing, because the whole slab of grid
blocks appears to melt at once or in a shor-spaced series,
This effect has also been noted by at least one other
author, with a similar explanation,
= Oil Rate Spike so
An
i i
* 000,
OO ee ake a
Fig 19-50 Slabbing—Oit Rate SpikePractical Reservoir Simulation
Detailed SAGD Pro’
cess Description
Steamflood front mechanism the mobile material and a gradation of saturations. These
Extensive work on these issues has been done diagrams are shown in Figures 19-53, 19-54, 19-55,
With respect 10 both water and steam flooding, with 19-56, 19-57, and 19-58
Sood matches obtained between analytical and lab-
oratory experiments, An example of this is shown in
Figure 19-51.
Historically, steamflood work
has concentrated on the interface «00 :
between the oil and the bitumen. It Loi iy Sasgueares |
is based on the assumption that the
steam will condense at the cold
bitumen front, This process would
| 3) farensenras
—-L5)
Tore fomn} |
sreaw F000.
be dominated by watercil relative
permeability curves. Behind this
Tewrenatune °F
3
region is saturated steam, and this
process will be affected by gasbitu ol
men telative permeability Iti tikely os
that the process in a SAGD chamber
involves recovery both at the inter-
face and in the steam chamber dur
worwaréariogs
Tit P |
ee
DISTANCE ALONG CORE - Fer
Comparison of calculated and experimental temperature proties for
laboratory mode!
ing a period of time. —— TO
Convection
| steam SOLUTION Gas
ancantimouan
In the strictest sense, convection
means the ansfer of heat by the
motion of fluids that have a heat
capacity. In classic steam flooding,
the movement of steam behind the
front is a convective process.
The context for this discussion
about convection concerns the
movement of steam inside a SAGD
chamber This was clearly envisaged
by Butler, who, in his theoretical
developments, shows material mov- 1
ing down the interior of the cham-
ber as shown in Figure 19-52
This is observed in reservoir sist °5
ulation. It was first identified in the
literature by fto et al They show a Comparison of
series of diagrams that illustrate a
progressive change in the depth of
ot necoveny -Pv
‘OAD On
steam F900:
Ta 20°F
S hexrennwenrac
= carcuraTeo
7 @ W820
FLulosArooucto - Py
calculated and expenmenial ail cacavery curves for
laboratory moce!
Fig, 19-51 Steam Flooding AdvancePOSITION OF =
INTERFACE) Su
Tat N
STEAM /
CHAMBER ¥
a
NORMAL oniGiNaL
veLoerry RESERVOIR
ve
~ near
CONDUCTION
Fig. {9-52 SAGD Theory Developed by Bauer
cs ot oe anaes
Fig, 19-53 Simulation Reswitstajacentt0 Steam Chamber
‘This seems to have caused a smal furor, The author
sees no conflict between the theoretical development
and the results of simulation. The equation developed by
Butler is shown as follows:
20A5,k, gah
mv, (194)
ermal Reservorr Simulatio
Fig, 19-54 Deailed uotherms of SAGD Simulation
Fig, 19-55 Detailed Contours of Oil Saturation at Edge of SACD
Chamber
Details of this gradation in saturations are not specif
cally highlighted in Butlers theory which uses some
clever integration and assumed properties. The assump
tion in question is oil permeability—ie., the product of
,, and absolute permeability He did account for the vari:
alion in the viscosity of the oll with temperature gradients
in the mobile zone. He solved for oil flow, since he was
principally interested in predicting production rates,
Water flow in the mobile zone did not appear to be ofPi
tactical Reservoir Simulation
Fig, 19-56 Detailed Steam Flow Contours at Edge of S260
Chamber
‘ae ag”
Fig. 19-57 Detailed Water Flow Contours at Edge of SAGD
Chamber
Fig, 19-58 Detailed Oit-Flow Contours at Edge of SAGO Chamber
particular interest, and he presents no calculations, He
must have been aware that water (steam condensate)
was flowing, but, normally, water is not sold.
The results have been interpreted to indicate that the
rate is Independent of the shape of the interface. We
know, for simulation, that SAGD recovery efficiency is
heavily affected by rates and operating conditions. This
also affects steam chamber shape, which will be dis-
cussed in more detail later in this chapter. There were, of
course, some justifiable simplifications made in the
development of the theory
The author has discussed some of the simulation
models directly with Dr. Butler. One of the problems he
observed was that there is a discrepancy between the
rates and permeabilities predicted from his theory and
those predicted by simulations. Numerical errors in the
simulations were suggested as a possible source of this
discrepancy. Having done a number of grid sensitivities.
the author does not believe the problems are numerical
but concurs with the comment about permeability: The
author’ explanation is somewhat different in that geome-
chanics indicates large permeability changes near theite and more imederate changes away from the
bilities, the two methods likely pracluce substantially
sinuilar results
Operating conditions
Operating conditions provide some interesting impli
cations, Referring to Figure 19-59, the steam goes in the
upper horizontal injector condenses at the
n is produced out of the reservoir front the lower
terface
horizontal producer. E portion of
a cettain
the heat in the reservoir is simply produced and does n0,
useful work, The amount of heat that is transferred to the
cold bitumen is going to depend on the temperature dit
ference between the cold
the amount of heat that is swept away through the pro
ducer, More heat should be transferred with slower side
chamber velocities and t
a thinner zone would require less material to travel
re steam and biturnen and
her temperatures, Conversel
that the objective of pro.
The
ot totally Wasted
that is produced from the producing steam is
The heat cant be (and is) recovered in
A beter
steam oil ratio
heat exchangers, This changes optimization
measure of efficiency than the cumulativ
SOR) would be:
oil prod
heat in ~ recovered heat out) divided
Solution gas
Most early simulations done in cyclic steam were
ad
3
done with no solution gas in the oil, ft has been arg
that, at low solution gas oil contents of 4 m'/m*
selfbbl) or less, solution gas drive will he minimal
Further itis likely that cold bitumen is sufficiently viscous
that yas bubbles will not form 2 continuous gas saturation
in the reservoir and lead to gas relative permeability
Optimum Drawdown
intertace at wailbore
Too litle drawdown:
‘Too much drawdown
steams produced
Fig, 19-59 Heat Loss Due to ProductionPractical Reservoirs Simulation
Layers of gas are found in some heavy oil reservoirs
and this has suggested a number of things
‘The layers mnay indicate sealing barriers within
the reservoir
The gas should segregate,
Gas sealing, as outlined eatlier in this book, is con-
trolled by capillary imbibition barrier pressures, This also
varies tor wetting and nonwetting phases. Consequently
even though there are thin traps of gas within an oil zone,
this does not imply that there are seals to steain or that
increases in reservoir pressure due to steam injection
won't cause the gas to leak
From the perspective of segregation, geological time
is evidently available for the in siw conditions. The
author has heard that solid residues from heavy oil wells,
will remold if let in a beaker for several days. So, while
bitumen can be near solid, its not static on the scale of
several days. Clearly reservoir temperature would affect
how quickly segregation would occur To the author’
knowledge, there are no tests on this issue on the pore
scale level
Having said this, reservoir simuiation model results
are affected by the inclusion ofsolution gas. A number of
these issues with heavy of relative permeability were out-
lined earlier This issue remains unresolved.
Foamy oif flow
Foamy oil was originally proposed by Dr Gerry
Smith. His proposal was based on pressure transient test
interpretations of heavy oil producers and was based on
fa logical argument that there must be some viscosity
reduction to account for interpreted. permeabilities.
This is not a simple issue and includes the effect of
wormhole propagation and the attendant increases in
formation permeability
It has also spawned a great deal of fundamental
research on oil Now in reservoirs. Some of this work has
included tests with glass micromodels to view what is
happening on the pore scale. The most recent research
has indicated that foaming does not appear to be occur
ring in glass micromodels.
In the authors opinion, the permeabilty is changing
more than the liquid viscosity and this is leading to the
high permeabilities on primary heavy oil production.
However, more research is required to categorically prove
permeability increase over viscosity decrease (foam
flow). It appears that the foamy oil research is more
mature than the geomechanical analysis.
or
Thermal expansion
Thermal expansion occurs when the oil sands are
heated, There are a number of implications to this:
Tre fluids often have a higher coefficient of
thermal expansion than the solids (sand
grains). In cold bitumen, which has a very high
viscosity this pressure will not be able to dissi-
pate laterally and vertically To some extent, the
sand will explode Butler analyzed expected
pore pressure increases and dissipation time in
2 paper entitled, The Expansion of Tar Sands,
During Thermal Recoven’™
lf the stearn chamber expands, the expansion
and stresses must be absorbed vertically and lat
erally Thus, the cold sand between chambers will
be squeezed horizontally At the same time,the
vertical expansion of the steam chamber will
cause the overburden to be “jacked up" to some
extent, which will decrease vertical stresses with-
in the steam chambers. During the early periods
of heating, there will be concentric expansion
surrounding the injection and production welt
for a SAGO pattern.
In effect, the differential expansion of oit sands fh
ids and solids will reduce the effective stress or reduce
sand grain contact pressures. This will increase perme:
ability somewhat, but it will enhance the tendency of
the oil sands to shear The differential expansion
between the cold and hot areas will cause shear stress-
es, and this wifi facilitate shearing, The latter greatly
increases permeability.
In the literature, there are examples of stress path
changes with time. Timing is very important. The etfec-
tive stress decrease originally occurs at temperatures
before the bitumen becomes mobile. The stresses from
expansion are absorbed aliéad of the steam chamber in
the cold bitumen. Therefore, shearing should occur in
the cold bitumen and within the steam chamber and
cold bitumen transition.
Shearing
‘The general concept of shearing is shown in Figure
19-60. in essence, the stresses induced by the steam
chamber (both fluid pressures and thermal expansion)
cause the oilsands to ail. When this occurs, the perme-
ability is enhanced as described earlierfhermal Reservoir Simu
Reduction in Mean Effective Stress due co Increasing Pore Pressures
sy
Phase |
Fig, 19-60 General Concept of Sein
The Phase A pilot at AOSTRAS UTF project was exten:
sively instrumented. Piezometer readings within the
Phase A pilot clearly demonstrate that pressures were
being elevated to nearsteam pressures well in advance of,
the steam chamber, which is indicated by the presence of
@ temperature increase. Another example is available
from the JACOS Hangingstone projects. This data is,
indicative of fluid mobility within the cold reservoir out
side of the steam chamber. These elevated. pressures
reduce the effective stresses within the cold oil sands,
which encourages the tendency toward shear failure
Required permeability changes
Based on the early UTF modeling, it would seem that
very high permeability increases and porosity changes
are required throughout the reservoir. However, if one
considers a steam chamber, reservoir permeability has
the most effect in the immediate vicinity of the well
where there isa flow concentration. This is potentially sig-
nificant. Large dilation effects may not be required
throughout the entire reservoir to achieve production
increases, It also suggests the possibility of stimulation,
ARE conducted a sensitivity by comparing two cases:
a reservoir of high uniform permeabilities of
7,500 mD horizontally and 2,500 mD vertically
(nearweltarget.dat)
+a reservoir of undilated permeabilities of 1,775
mD horizontally and 1,509 mD vertically (near-
wellbase.dat)
A third run was tuned to match the production pro-
files and cumulative recovery, The results are shown
on Figures 19-61, 19-62, 19-63, and 19-64. Indeed,
the permeability changes required are far less drastic
than is suggested by modeling air permeabilities (or
the entire reservoirPractical Reservoir Simula
The maxiniuin increase in permeability isa factor of 2
in the nearwellbore region, It is appropriate to consider
the effects of drilling and completion also, The well is
drilled to a larger diameter than the liner. Most projects
Use either slots or a screen to prevent sand production:
therefore, the sand falls onto the slots or screeds. In this
lation comes from reservoir engi
case, not all of the
‘neering effects. This occurs directly adjacent to the well
‘where it will have the most impact.
Away from the wellbore, a 30% increase in perme.
was required. The modeling showed that no tea:
could
sonable near wellbore alteration in permeabili
account for the production increase required. From this,
ARE concluuled that geomechanical effects must have
‘occurred al, or ahead of the steam chamber interface.
The results are unlikely t6 be unique. The changes in per
meability are unlikely to be as abrupt 2s was history
matched. A transition from various levels of permeabil-
ty could be incorporated. However, there would be no
OU Rate SC (msdday)
basisto determine which of these solutions would be the
results show be viewed! as approximate
lly, laboratory measurements coukd be made of
Athabasca oi sands to directly determine these effects.
In practice, obtaining undisturbed cores is both expen
sive and difficuk,
With an in sita porosity of 28% and an index of dis.
turbance of about 30%, the nearwell pair porosity
would increase to about 36.4%. Based on pressure tran:
sient analysis the undisturbed permeability of the for
mation is about 2.995 mD. Aa increase in permeability
by a factor of 2, due to disturbance, indicates the near
well permeability requised is about 5,850 mD. Th
within 20% of the average air permeability derived from
xt, the previous history-match perme-
core. Ia this co
abilities are logical
Figures 19-65, 19-68, and 19-67 stow the perme.
ability distributions and the shape of the steam chamber
derived
a TT
Time (Oate)
Fig, [9-61 NearWel Permeability Enhancement—Oil Ratesmal Reservoir Simulation
oo 2008 20s 2008 aor? 3008
‘ime (Oat)
Fig, [9-52 NearWell Permeabiliy Enhancement—Injection Rates
‘2002 2003~=~*« SCS: SCOOT SSC«O0B~SCSC«OCSC*« TSC
Te (Cat)
Fig, 19-63 NearWell Permeability Enhancement-—SORaoe 26083068 2007
Time Oat)
Fig, 19-64 Nea Wel
Permeability Enhancement
Fig. 19-65 Permeability Distribution in
Uniform High Permeability Reservoir
20082008 ==« 2010S
rmalative Oi! Production
Fig 19-66 Permeability Disribution to
Match Use of Air PrmeabiliesFig, 19-67 Ternary Diagrara of Steam Chamber
‘with Lovalized Permeability Eahoncernent
Hydraulic fracturing
Toe possibility of
SAGD has been debated
fracturing, for
into two possibilities:
the use of hydraulic fracturing as a startup
@..t0 establish communication
et
the injector and producer
the use of hydraulic fracturing as a generat
injection technique
At present, the concept of injec fracture
ily due to concerns about
pressure is not popular prim:
containment above and below the pay zones atid poss
ble distortions to the steam chamber. It seems to have
worked in eyelic steam stimulation,
The first possibility could shorten the startup time
dramatically: Thermal convection is a slow process,
and reducing the 3 to § month startup time would be
economically advantageous. The AEUB has allowed
this to proceed on AEC’ Foster Creek project. In the
author’ opinion, this is a realistic application: howev
cer, time will tel
Typical Issues in Bitumen-Prone Areas
Reservoir characterization
Invariably, distribution of permeability and porasity is
{a major issue in reservoir simulation. The two largest
areas of biturnen deposition are the northeastern section,
of Alberta and the Orinoco belt of Venezuela. The best
pay is usually found in areas with stacked fluvial chan-
nels. This is ue of the Kern River field in California, also,
Channel environment
Fluvial environments are characterized by rapid lat.
eral changes, Experience has shown that channels have
characteristic dimensions. For instance, as shown in
Figure 19-68, itis possible to estimate the length of river
meanders based oa modern geomorphologic
studies 9 For the internal structure of fluvial channels,
2 typical width to thickness ratio is 10:1 (Kuplersberger
and Deutsch).*
Channels or, more correctly, point bars have charac-
teristic cross sections An example of this is shown in
Figure 19-69.
There are a number of key features. Note that the
point bar accretes on the inside of the bends and grows
to the outside of the bends as shown in Figure 19-70.*
lethe river goes through an annual cycle or has period
ic flooding, then there will be deposition of different
energies and particle sizes. In the final stage, when the
channel is buried, the watercourse that is shown on the
right of Figuee 19471 can be filled with either perme
able or impermeable material This is known as an
abandonment plug
A umber of discontinuities are present, which are
usually muds or shales. To summarize, there are three
kinds of mud relevant to production:
Inwerbedded muds
Inclined heterolthic stata (IHS)
Abandonment plugscal Reser
i
| A noawt
| A247 28
| (eopo.a ans Warman 1360) | ff] tonene
| | Rn
| |
| /
| Mangere wisth=i6 10 29. times | coer
tn I tasiane_ 18721
imtSont ; +
Fig. 8. Schematic eprasantation of large scale eoint bar
based on channel width and meander wave length (from Fl
Mossop, 1985)
Fig, 19-89 Characteristic Coss Section
Fig. 19-68 Characteristic Dimensions of Channels ‘of Fluvial Channels
Horizontal Well Vertical Well
CHANNEL,
FILL
— > Accretion Direction
Fig. 19-70 Development of Point Bar Depositson Ape HOB
Fig. 19-71 Abandonment Plug—Foint Bar Deposits
Geostatistics
Geostatistics is a highly appropriate tool for analyzing
the distribution of these different facies; however, such
descriptions must be properly tuned. This involves
extremely detailed description of rock properties and
geometrical distribution. From the point of view of geo
metrical distributions, ARE has not been able to find
other analogues to the McMurray in the Athabasca area,
The work of Strobl et al, provides some insight, in par-
ticular with respect to IHS; however, this is insufficient for
the purposes of quantitative description." More detailed
outcrop studies could provide useful base data, There are
a number of river outcrops, and data is available from sev
eral surface mining operations. It will take years of
research to produce quantitative results,
‘There have been some attempts to use data from
Prudhoe Bay and the Tillamook estuary in Oregon; how-
ever, all of these have significantly different morphology
and scales. The use of inappropriate analogues does not
provide ary useful insight.
Fractal geostatistics were used for the UTF sites, as
described in Muktierjee et al. Due to computational
tations, the results of a number of individual pattern mod:
els were concatenated, This is shown in Figures 19-72
and 19-73.
Figure 19-74 shows a method in which the distribu
tion of mudstones was broken into smaller scale models.
Adding the results of these individual runs permits one to
perform a detailed large-scale model. Note that width of
the element of symmetry is 35.0 meters. This corresponds
20 a coral pattern width of 70 meters.
Discontinuous barriers in SAGD
In conventional oil reservoir engineering as well as
simulation, the effect of these interbeds has been quanti-
tatively calculated. The fist step was to determine statis
ically tne shale distribution as discussed in chapter 4
The major reservoir performance effec is to reduce vert
cal permeability due to increased flow path lengths. This,
is shown in Figure 19-75, Note that these calculations are
usually done with singlephase flow.
Returning to the previous model prepared by
Muhkerjee, the fractal model results in a much less con-
tinuous reservoir description. Ths is important for SAGD.
simulation since continuous shales form barriers to steam
rise, whereas a series of smaller shales with gaps in
between will present a minimal disruption to steam rise
This is shown in Figure 19-76.
> 418Practical Reservoir Simulation
UIT
cons
—eS
ean)
cone
MCOSL USING FRACTAL OISTRIBUTION
g™ ge oe
SS
—
aT
Fig. 19-12 Disinbution of Barriers in UTF Modeling —Vertical Cross Sections
as
Fig. 19473 Distribution of Barriers in UTF Modeling —
Arrangement of Cross Sections
usr 7
on
It anay be coneluded that the effects
of barriers in SAGD involve fund
mentally different processes and
physics than in conventional reservoir
engineering, This issue has been
approached Irora an experia
spective also
Lab testing of mudstone gap
size on SAGD performance
Yang and Butler did an extensive
laboratory study on this issue using
scaled lab experiments. They examined
a number of interesting geometries."
The first involved the use of a plas-
tic barrier that extended from immed
ately adjacent to the well all the way
across the pattern. This is shown on
Figure 19-77. The results showed that
as long as the heat could transfer
upward and there was a small gap, the
bitumen would be drained. By analo-
gy, a very small gap in the mudstones is
all that is required to provide good ver
xical communication
The second case involved the use of
a plastic barrier that extended from the
well all the way across the pattern with
only a gap at the edge of the pattern.
The results of this experiment are
shown on Figure 19-78. In essence, the
steam migrated along the bottom of the
plastic barrier and then grew upwards
from the end, Following this, the bitu-
men started to drain away from the wel
con top of the barrier and then reversed
backward toward the well underneath
the barrier
Both of these experiments are
strong indications that, while discontic
uous mudstones may delay oil recov
ery, they do not impair ultimate recov
ry. It also establishes that even a small
gap near the wells will have a major
elfect, A gap that occurs most of the
way actoss the pattern will eventually
provide drainageThermal Reservoir Simulation
Fig. 19-74 Method of Combining Results From Multiple Models
Indes proper of images
[os [earen ae um]
rap aes fs eo
wis-[a6s [2] 199 10
wa aene | os | 391 S10
gia [ a6 [6s a re
336 _[ 078 -[ “397s
eis-[ gis) ss) se
‘= oa suber of paris,
1,» sector mean econ (Dears)
Yat svete sage (0.
(p= oli lee of igaoacs,
am,
tao az,
ar SEB
Cam =e
"ena acne ‘taal pn)
Flow pats peependieslar aod parallels grin azar
Depletion
Fg 09
5 Increased Flow Path Lengihs—Single Phase Fig, 19-76 Growth of Steam Chamber around Barrer
> 48Fly. [9-77 Physical Flow Model of Yang and Butler
Reservoir simulation of UTF
and Hangingstone
More recently, to, Ichikawa, and Hirata have
shown that steam will detour around. low
permeability areas, This is shown in Figure 19479
which illustrates the results from reservoir simulation
and a vertical array of thermocouples. Further results
are shown in Figure 19-80."
Ito etal. state further that low-permeabilty zones and
such phenomena were seen i seven out of nine obser.
vation wells in the UTF Phase 8 area, which comprised,
Fig, 19-79 Growth of Steam Chamber ret
Stearn Flow around a Barrier
Fig. 19-78 Cross Section of tsothers ant
Flo
ctors qnound Lath Plastic Barier
three injector/producer pairs. The diameter of the
detours was calculated to be less than 10 meters in most
cases. Based on this rate, 775% of wells will have some
form of barter in them,
Since the fundamental geological environment of
the McMurray sands is similar, it can reasonably be
extrapolated that the majority of bitumen wells will
have some kind of barrier present. Large sections of
pay, without any bartiers whatsoever, are the exception
for SAGD development
Fig, 19.80 Siraration Cross Section of lsorherms
‘and Flow Vectors around BarrierKern River shales
Similar experience has been demoy
strated in the Kern River field. Figure
BSL depicts the shale distribution
determined using detailed descriptions
for this field.” Key conclusions from this
study are
Detailed heterogeneity modeling is
required and can result ir yood
performance matches.
Discontinuous shales allow signifi
cant oil drainage from upper to
lower sands As a result, ower
sands have higher apparent recov
eries and upper sands have lower
apparent recoveries,
‘Small pattern element or single
sand models are inadequate to
predict recovery.
Properties of mudstones
“The shales found in the unconsolidated sediments of
northeastern Alberta are not shales in the context of con:
ventional weservoir engineering, In fact, the progzerties of
shales vary considerably.
For instance, as a junior engineer, the author found
many of the log analysis examples from the Gull Coast
did not apply to Alberta. Gulf Coast shales are much
cleaner and (ypically have resistivities of 2-3 ohms. [i
the Western Canadian Sedimentary Basin, shales typical:
ly are more a mix of silt and clay and have resistivities
that are typically higher, usually 8-10 obms. One trick in
log analysis is to use the resistivity of the overlying or
underlying shates to estimate shate resistivity: The inher:
ent assumption is that the clay in the sand is the same as
the shales above and below
The question then becomes, Is the shale in the sand:
stones pure clay or a mix of elay and silt? strong argu
ment can be made that the pore sizes are sulficiently
small that only the clay portion ofthe overlying sands and
shales is really in the reservoir rock. Therefore, shate
adjustments should be done with clean clay resistivities of
2t03 ohm-meters instead of 8 19 @ohm-meters. The con-
verse argument is that the clays are intrinsically ditfer
enti. there are many clay minerals with different con
local shale, realy clay and sil
ductivitiesand that
should still be used,
Fig. 19-81 Stochastic Distnbution of Shales in Kern River
The truth is somewhere in between, The clays are dit
ferent. but the shales above and below are not clean
s. The trick that can be safely used in the Gulf Coast
reeds t0 be applied with caution elsewhere
The mudstones in uncemented formations are quite
ditferent from pure clays, They are, in fact, a mixture of
clay. silt, and sand. There is considerably more sand than.
is typical of most shales and, as a result, the properties of
the mudstones are not typical of shales. They have con-
siderably higher permeability and porosity {a this regard,
there is a substantial deviation in the pattern of reservoir
performance that one comes to expect from convention:
al reservoir experience.
This cannot be determined by looking at logs. It
requires detailed geological descriptions. Thin sections
have proved to be very useful in understanding the
behavior of these rocks.
Layering
The effect of discontinuous barriers has been dis
cussed in some detail. Like almost all other reservoirs
the fluvial channel environment also has layering and
interbedding.
ARE performed some studies using a description
based on logs from an actual well. The model included
interbedded sands and silts immediately above the injec:
tor and midway through the section. The largest elfect
‘was to cause the shape of the steam chamber to widen
This is the opposite ofthe upside-down triangle predicted
by Butlers theoryInterbed
Interbedaling is a series
f alterna
1g eaudistones auc
sands. This reservoir is clearly of lower quality than the
massive cross-bedkled sands that are present in channels
Experience at the UTF indicates that the steam chamber
slows in height growth in these sections, This was sum
marized in an averview context by Faroug Ali
Shape of the SAGD steam chamber
To date, the most comprehensive data regarding the
actual shape of the chambers comes from the UTF site
The plot shown in Figure 19-82 does not show the cl
upside-down triangle predicted by Buller from theory
that is demonstrated in many lab experiments The the:
ony is not wrong per se, hut actual reservoirs are rarely
omogeneous
The author has dubbed this the manta ray shape,
since the chamber undulates somewhat and is pointed at
the edges. There has been some intense discussion at
ARE a8 to whether this shape was caused by geomectian.
ical effects. Overal, based on ARES geomechanical runs
geomechanics do not appear to be responsible for the
shape of the chamber, Inthe author’ opinion, the shape
of the chamber is controlled mosty.by geological effects
such as layering and discontinuities
ARE did a number of sensitivities with the shape of the
chamber Inthe festcase, we input alternating layers with
permeability increased by a factor of 2 and decreased
(divided) by a factor of 2. The porosity was left constant
This is shown in Figure 19-83
Fig, 19-82 Shape of Steam Chamber rom the UTF
19-83 Alternating Layer Sensinuie
In the second case, we input a gradation of perme-
abilities as shown in Figure 19-84, This was expected to
give the manta ray shape. The actual effect was to change
the shape of the chamber into the shape of the superman
crest with the truncated upper edges of the upsidediown
le rather than a manta ray
In the third case, we input a series of
discontinuous barriers, The pattern is too
dense and too closely spaced to be reali
tic, The shape generated certainly indi
cates an erratic development as showa in
Figure 19-85,
Inthe final analysis, it seems likely that
combination of Case Two with grada
tional permeability changes, Case Three
with some discontinuities, and the
Layered Case with interbedded layers near
the horizontal wells (which produced the
rightsideup triangle) could be combined
to match the shape observed in the UTF
Phase A. Note that there was an inter
bedded layer in this pilot area.
will be interesting to see further
results as they become available.
trianFig, 19-34 Gractation of Pacmeabities
Tracking the development of
the steam chamber
itis easiest to track the overall development of the
steam chamber by monitoring the progres of isotherms.
This change is quite dramatic. However. there is some
conduction away from the chamber and a gradation of
temperatures. Therefore the change of colors onthe out
put plots does not indicate the precise location of the
steam chamber interface
‘The ternary diagram indicates where the steam
chamber itself is located. The ternary diagram is a
colorcoded triangular plot, which indicates whether
the fluids in the formation are water, oil, or gas. The ter
nary diagram cannot indicate if the gas is steam or
hydrocarbon (methane)
‘The pressure plot shows a general increase in pressure
within the steam chamber. However, pressure increases
precede the steam chamber. The pressure increase in the
reservoir is caused by a number of occurrences. The oil
expands with temperature, as does the matrix. If the
amount of thermal expansion is ditferent forthe matrix
and pore fluid, then the pore pressure can either increase
Fig. 19-85 Discontinuous Barriers
or decrease, The reservoir fluids can move to equalize
these pressures. Until high temperatures are reached, the
bitumen viscosity is very high, As a result, these pore
pressure increases dissipate slowly: Pressure increases
can occur from stress increases also,
Water above and below
Generally speaking, water below is not thought to be
a problem in SAGD. Some vertical standoff is required,
but this is in the order of | t0 3 meters. This may leave a
‘small basement of oil that is unrecoverable.
‘Top water can also affect recovery When the steam
chamber rises vertically into the water zone, the water
will segregate by gravity. The water 2one also absorbs a
Sreat deal of heat, A tiny skiff of water on top of an oil col
umn will not be significant. Therefore, it is the relative
amount of water and oil pay that is significant. A general
rule of thumb is that up to onesixth of the total pay can
be water filled without noticeable effects on production
performance. At proportions larger than one-sixth water,
itis matter af economics
> 40sPractical Reservoir Simulation
Gas above and below
Gas below may seem counterintuitive. In fact, the gas
isnot truly below—ie., below the oil pay—but rather dis
tributed in layers through the pay. As discussed previous
ly thin layers of gas do not necessarily indicate a barrier
to steam oF higher injection pressures, unless the uct
stones are quite thick, Gas layers are more common in
the western part ofthe MeMurray bitumen trend
Gas above the oil pay also affects performance. So far.
most ofthe efforts regarding this issue have been directed,
to estimating the effects of gas cap pressure reduction
caused by gas production, This has resulted in some of
the bitterest controversy ever seen in the Canadian indus:
try The Alberta Provincial Government fas held three
hearings on this to date:the General Bitumen Inquiry. the
Surmont Hearing, and the Chard-Leismer hearing. No
decision has been rendered on the last hearing, The first
two are available on the Internet at the AEUB website
‘They are in a pdt file format and are free. Rather than try
to paraphrase these comprehensive reports, the author
recommends that individuals read them directly,
Implications of thief zone
From a SAGD design perspective, the thief zone rep-
resents a serious technical problem. The steam cham
ber must be run at lower temperatures and pressures to
prevent the loss of steam, Steam loss cannot be pre-
vented entirely. This can represent a real loss in efficien-
cy, which translates directly into the potential econom-
ies of the project
Thief zone
The top of the Wabiskaw/Upper McMurray sands are
thought to be in communication on a regional basis
‘This can be represented in a number of ways. One way
is to use a dummy well to remove high-pressure fluids
Another way isto extend the grid with increasingly large
rid blocks. Since there is no steam chamber advance
into bitumen in water and gas zones, larger grid blocks
can be used.
ARE has prepared a number of such sensitivities using
both methods. The initial sensitivities were done with a
half well pair pattern model. Early runs withthe sensitiv
ties indicated high steameil ratios and high heat losses
fora single pattern. These results were, however, not real
istic. Since the actual projects are implemented in pads
with mutipte wells, modeling was changed to include a
2’epattern arrangement, On a proportionate basis, steam
losses dropped considerably and reduced steamoit
ratios. The author recommends that thie! zone tosses be
modeled. as best one can, with elements of semmetties
'spical of actual development
The production of gas has introduced some signi
cant pressure gradients in the overlying upper sands
Restated, the thief zone pressures are unlikely to be cont
stant overlying a potential SAGD development. If this is
the case, there may be influx from the boundaries adja
cent to higher pressured sands and losses atthe bound
aries adjacent tothe lower pressured overlying sands. The
combined steam chamber from multiple patterns may be
expected £0 equalize due to the much lower viscosity of
steam than water.
One potential advantage of modeling the reservoir
away from the pattern is that overburden heat losses can
be accounted for We used null grid blocks for the eser-
voir outside of the pattern—with an extra margin. The
underburden heat losses in the overlying thief zone
away from the margin were not calculated. In the end,
we felt that all ofthe heat was lost through the dummy
well, and this was probably a more conservative
assumption. In reality. a new pattern might be built
eventually next to the existing pattern, and some of this,
heat might be recovered.
This suggests that there may be considerable merit to
attempting to block the permeability in the regional
sands adjacent to a SAGD project. Experience in con-
ventional oil and gas suggests that implementing this
will nt be easy
Field thief-zone losses
Field data is available from one project at Kear!
Lake in Alberta, This was a pilot project where there
was an overlying gas zone. The pressure losses here
reduced the efficiency of the steam injection. Natural
gas injection was used in a circle around the pilot well
to repressurize the area, Pattern containment is a clas:
sie pilot project problem with conventional water-
flooding. Since the patterns do not repeat themselves
to provide containment, it is well-known that pilot
responses typically underestimate field scale water-
flood implementation. The increased pressures that
occurred due to gas injection did improve production
response; however, a considerable amount of injected
gas was required.From an operational perspective, it will be difficult to
evaluate the exact operating pressure without knowing
the thiefzone pressures, We have not iound any way to
detect breakthrough easily from the simulations tha
hhave done. Gult has indicated, in a government applica
tion, that they feel thae water cut te when this
will in
occurs. Giveg the problems with relative permeability
determination, outlined previously, the author is skept
cal. Experience will help resolve this issue; unfortunately,
such information is not in the public domain yet.
SAGD Design Issues
Criteria for evaluating performance
The major criteria for evaluating SAGD are:
+ Breakthrough Time. This is the time that it takes
for communication to occur between the hori
zontal injection and production wells. It isa func.
tion of input heat as well as the distance between
the injector and producer.
+ Production Rates. This aftects economics directly.
Higher rates translate into revenue. lt is common
practice to refer o the maximum rate, because
this usually indicates the overall level forthe pro
duction forecast.
+ Recovery Factor. Tis is an indication of the total
amount of resource that can be expected to be
recovered, However a high recovery factor does
not necessarily translate into the most efficient
economic recovery
‘+ Steam-Oil Ratio (SOR)/Oil-Steam Ratio (SR).
‘This is an indication of the efficiency with which
bitumen is recovered. It can be specified on an
instantaneous and cumulative basis. Normally the
Cumulative SOR is used. tt inversely describes the
overall efficiency ((e.,a low SOR indicates an eff
cient SAGD process). Usuallythe instantaneous
SOR is quite high atthe beginning ofthe steam
Injection process. This is inherent in the physics of
the process, However, toward the end of the steam
injection process, the instantaneous SOR can be
‘an indication of when the project should be ter
inated economically
+ Rise Rate. Many people use rise ate as an
indication of SAGD chamber performance.
This is normally expressed in em/day or
in./day The author regards this as potentially
misleading. Chamber shapes are not necessari-
ly the classic upside-down triangular, and dls
comtinuities, even of significant dimensions,
can affect rise rate without really reflecting
whether the SAGD process is working
However, given a reasonably homogeneous
reservoir, this can be a useful indicator,
In the near future, the author believes that these crite-
ria will be changed somewhat. This will be discussed
under global optimization,
Well length
Once the curm has been made in a horizontal well, the
cost of a horizontal well does not appear to vary greatly
with length, This srongly suggests the use of longer well,
Friction pressure loses in the steam injection well can be
significant. Optimal chamber development suggests that
the steam chamber should be as uniform as possible
Detailed well models have been developed for modeling
the flow in horizontal SAGD wells, The general result is,
that large casing sizes are used for SAGD wells. Typically
1778 mm (7 in.) casing seems to be the most popular. In
some cases, 244.5 mm (9% in.) casing is used. This sug-
gests that there will be a point beyond which longer wells
are not advantageous.
Since $:D modeling is so time intensive, the author
has seen relatively little work in the literature on this
issue. Intuitively, one would expect that as heterogeneity
increased, steam chamber problems would increase.
There is not yet much practical experience with this
issue. We do know that the fluvial environments have
rapid changes in reservoir properties and, in the author's
estimation, this will favor the use of shorter rather than
longer wells
ermal Reservoir Simulation
437etical Reserveir Simulation
‘There are practical comic:
eratines ice the use of fonger
horizontal wells. The chance
fof workover failures anc com
pletion failwres will increase:
‘with teat. Undoubtedty, one
of the more important aspects
in the horizontal section is
physical friction, Sereens asso-
ciated with slotted liners are
physically delicate
teappears thatthe best well
length is not known yet, The
fongest wells proposed seem to
be about 1000 meters (3250
feet), and the shortest are about
500 meters (1.500 feet). From
an economic perspective
ignoring operational problems
and heterogeneity, it would
seem that the longer wells
would be more economical
Based on experience in the
conventional side of the business, the author has used
short well lengths in simulations, The average seems to be
about 750 meters (2,500 feet)
‘Cumalatve Sten Ott Rath (in?)
Well spacing
More work has been done on the spacing between
wells. The best summary on this issue is the work of Ito
Ichikawa, and Hirata, which is shown in Figure 19-86.
This work indicates clearly that there isa link between
recovery and well spacing, This will depend on the height
of the formation also. It seems that most designs for com-
mercial projects are in the range of 70 to 110 meters
between wells, with 90 being typical, The author suggests
that this is an appropriate step for designing a fiett prof
ect in the detailed design stages.
Vertical spacing between producer
and injector
‘The effect of vertical spacing has also been studied.
‘This was examined by Edmunds and Gittins and is shown
in Figure 19-87, With larger vertical spacing, more time is,
required to initiate communication between the injector
and the producer. Ultimately this isan issue of timing and.
‘changes in rates.
so Ome
‘aan ek am Rw
Exapeed Tim ater alia Circulation (day)
Fig 19086 rect of Hortssaga! Well Spacing—Cumutaive SOR
2 . ‘ 4
Figure 4
SAGO Stanup Time vs. Spacing &
Potential Oiference (1 Darcy)
ig. 19-87 SAGD Startup Time vs. Spacing
‘and Potential DifferenceSAGD start-up
Most SAGD
have been periormed using 2-D cross sections. Actus
Simulations. as previously: discussed.
wells are never spaced evenly apart. In the past, detailed
ucted
studies have been cor sing the aetual direction
al surveys from the wells. True 3-D models are difficult 0
run since the longitudinal grid blocks are normally quite
large: they should be of fe same order of size as the hor
zontal dimension, Most of the comments in the liters
ture are quite oblique on this point. The concern seems
tobe to develop the chamber in a manner that is as even
as possible,
Giver that most conventional reservoirs demonstrate
permeability variations, as do oil sands reservoirs, the
author's expectation would be that chamber
growth would be more strongly dominated by reservoir
factors than the distance between wel’
‘The author has an example where long distances
were used, andl there isa section that apparently does ot
because:
dleveloy increased well spaciny.
©. The concern with this resu
is thar
the block is si wide that lateral lock
volume is not possible, This simulation, which is shows
in Figure 19-88, would nox, in the authorSopinion, actu
hypotherical examp
eating of the gti
ally represent what would occur in the resewoir
To mote correctly model along the length of the well
a finer grid should be used as shown in Figure 19-89, The
steam chambers now cnalesce along the axis, consistent
with cross-sectional mode
The data of Strobl et al
Figure 19-90 supporss the view that reservoir properties
affect chamber shape.” This is related to the IHS dls
cussed earlier and the rip up breccia in the lett corner.
The IHS, which occurs on a fine scale, is difficult to
model, This would not be a problem if permeability were
input as a true tensor, as discussed in chapter 2
from the UTE site shown in
Fig, 19-85 J-D Optimization Using a Coarse Grid along Horizontal Well Axis
ma
i
Re
Fig, [9-89 3:0 Optimization Using a Fine Grd along Horizoneal Well cisPractical Reservoir Simulation
ne we ve
Fig, 19-90 Tempenatre Profiles ana Etec of {HS Measured in UTE Reservoir
Artificial lift
Provided sufficient pressure is available, the SAGD
process, once started, will provide steam lift naturally
The flow of these fluids to surface is not straightforward.
Figure 19-91 showsa plot from the UTF Phase A and doc-
uments pressure and temperature conditions in the
riser” Recall that the UTF Phase A wells were actually
located in a mine, and the fluids did not return to surface
as they would in @ regular well
Edmunds points out that they did not operate the well
where there was intermittent flow and that these condi-
tions are quite undesirable. Certainly, the intermittent pro-
duction seems to have interrupted the surface facilities.
‘Slugging is hardly new to the oil business; this phenome-
non is observed on many conventional wells, Still, some
idea of how big and how frequent the slugs will be is
quite helpful, Sometimes these details are not transmitted
to facility design people, who may design facilities for
steady state conditions.
490 «
2e——
o —_
bs Peart _
3° TO ae
ae ees Te
o 2 ~ a le,
° b ft
“150 425 10 75 50.250
(BH Temp - Tea, 099 6
Fig, [9-91 Vertical Riser cho Mine) to Surkace vs. felet
Subcooting
One of the implications of Edmunds’ paper seems «0
be that intermixtent production is inbezently bad. This is
certainly true ifthe surface facilities were not designed 10
handle i. Conversely, Donnelly describes the intentional
se of intermittent flow for the Hilda Lake project.”Wellbore hystraulies prograins have been ttesigned
to calculate these effects, which have been called gey-
sering The analogy is to the geyser in Yellowstone
National Park in the United States. Donnelly compares,
the lift at Hilda Lake to a coffee percolator (Perco-Litt?)
Some of the results from the Edmunds paper are shown,
in Figure 19-92,
sonose
te sataly tg) (=
10500 { ‘
1080
Fig. 19-92 Btect of Steam Inlet Euthalpy on Vertical Flow in Wert
Indeed, lit does occur on an intermittent basis. This
program has been linked with a reservoir simulator to
solve both intermittent wellbore and reservoir problems
simultaneously. Butler has presented calculations using
steady state cortelations. This information may also be
used to determine lift rom SAGD wells. Butlers assump-
tions are somewhat different."*
The litt of oil can be aided by steam
injection during the early stages. No extra
facilities are required since all of the
required equipment is in place forthe ini-
tial startup phase. Some companies plan
to use natural gas as a form of gas lift
‘The use of gas lit can be expanded to
lower pressures by staging the lift. This
process, called E/ift, is shown in Figure
19-93 and was originally developed by
Dr Ken Kisman
Conventional suckerrod pumps are
limited by stuffing box rubbers and limit
ed rate capacity. The rates from most
SAGD wells typically make electric sub-
mersible pumps (ESPs) more appropri
ate, However, they are limited in the tem.
perature range across which they can
operate. Most people seem to feel that
Thermal Reservoir Simulation
FTP Celsius or about 800 kPa (340° Fabeenheit or 116
psi) is the limit of current technology ESPs also have
limited santé production capacities, This may change in
the fucure.
ESPs are more costly o operate and are more difficult
to put in the wellbore. At present, there is limited power
available in gortheastern Alberta, ESPs consume large
amounts of electric:
The cost and viability of artificial lt is a subject of
interest from an optimization perspective. Early optimiza
tion analyses have concentrated on minimizing steanroil
ratios. These attempts are quite preliminary and will be
discussed later
Geomechanical effects
‘The following information is required for geome:
chanical effects
in site stresses
formation strengths
permeability variation
In situ stresses,
In situ stresses have been determined for some areas.
There are a number of ways of doing this, The stresses
‘can be estimated from overburden calculations. In very
general terms (ignoring detailed tectonic stress analysis)
the largest stresses are horizontal at depths above 175
meters. In moderate depths, from 175 to 575 meters, the
> Ligue prseton
cond production wing
asing annus
fo sotomtole
et tion
om producton — Comph
Figure 9 ELit proucnoawelicomplenoe with ono m-hale pump in the sesond sage
Fig, [9-98 Bit Production Scheme
> arPractical Reservair Simulation
horizontal and vertical stesses are approximately equal
At deeper depths, below 575 meters, the largest stresses
will be vertical
Ieis possible to use mini-fracs to determine the mink
mum stresses experimentally This technique has been:
used in the conventional side of the business for some
time, These tests and data ate used for hydraulic fracture
treatment design. It 8 aso possible to estimate formation
permeability with such ests
These tests were applied at the UTF site, and the
results were published. The UTF site is quite shallow
and, as a rest, it is expected that maximum stresses
would be horizontat
It shouid be pointed out that horizontal stresses are
‘ten not uniform, There isa directional preference, and
this promotes failure.
This type of test and data interpretation is usually
done by specialist geotechnical engineers
Formation strengths
‘The degree of dilation that occurs is strongly
dependent on the physical properties of the reservoir
sands, This type of testing is not routine yet. Most work
has been done in the university environment,
Specialist geotechnical companies build and operate
testing equipment.
‘The most important aspect is getting samples that are
not disturbed. This was discussed from a reservoir engi-
neering perspective earlier in this book. Dusseault (1977)
examined the effect of sampling disturbance on speci
ren properties. Geomechanical properties such as shear
strength, stifiness, compressibility, and permeability are
very sensitive to disturbance
Rajani and Sanchez (1988) attempted to characterize
the geomechanical properties of the oil sands within the
heavy oil belt in Venezuela. They concede that sample
disturbance, as indicated by higher core porosities, had
affected their measurements of absolute permeability
and hoped that a relationship could be found to apply
similar results to in situ conditions. Chalaturnyk and Scott
(1992) described how reservoir properties could be
obtained from a geomechanical laboratory Reservoir
engineers were cautioned to use the appropriate stress
paths so that the laboratory tests would suit their particu:
lar recovery process. Vasquez et al, (1999) conducted
mechanical and thermal tests on unconsolidated sands
for the Tia Juana field heavy oil SAGD project in
Venezuela.” Core disturbance, when coring appears to
have resulted in disturbed samples, is indicated by high
492 <
er than expected core porosities, The resultant friction
angles of 25° to 27” appear to be low This was attributed
to sample disturbance
Getting good data is very difficult, and many
sophisticated attempts result in failure Therefore
acquiring good data requires some persistence, From
a practical perspective, the author has had to rely on
the bes! data trom offsets, which are often a consid-
erable distance from the actual field where work is
being conducted
Scott etal. (1994) described thei faboratory
results on Cold Lake oi sands. The volume and
permeability changes measured resulted from
changes in the effective stresses shear stress, and
femperatutes. In these cases, it iS necessary to bor
row from others’ experience.
Cldakowski (1994) studied the stressinduced py
meability changes of Athabasca oil sands at reser
voir temperatures (-8°C) and noted that,although
absolute permeabilities were in the order of 1,000
10 5,000 mD, the effective permeabilities to the
‘water phase were 0.01 mD Itis difficult to dis
place water around an immobile bitumen phase.
Shear stresses resulted in distortion and dilation,
which resulted in an increase in absolute perme:
ability if the confining stress was below a critical
level, Oldakowski found that dilatancy was pro-
nounced at failure and postfailure and that it was,
a function of bottr the confining stress and the
original porosity of the specimens.”
Chalaturnyk (1996) studied the UTF Phase A pilot
project. His laboratory program examined thermal
volume change, thermal conductivity,compress-
ibility stress-strain and strength behavior gas evo
lution, and the composition andi properties of oil
sands, shale, and limestone, Mere was an error
made in the loading procedure for the samples,
which caused somne disturbance elects
Accordingly. he did not use his own dilation
data. Instead, he used the data of Oldakowski."
‘The best data to date seems to be the data of
‘ouhidi-Baghini, These samples were taker
from the MeMurray in a river outcrop tn this
area, the McMurray is not bitumen saturated
This data includes a comprehensive suite of
tests It wit be discussed in more detall in the
{olowing section."Elastic Pl
8.0
6.04
Mote Coulamis = 10.MPa
1
5 7 Chim oN
tieAModel
See ; Ser. = HOWRa Model,
7
Thermal Reservoir Simulation
oma
~~~ Mone Covloms.<
Osa
Mone Couloms. 0
Mone Coulomb 0}, =428 MP3
= = = More Coutoma. 0 = 59 SP
2 1 Chalaums. 6 = 10 MP3
0.25 MPa Mose!
Creer es
akowsks ga 19 MP2
ttt h aaa aad
2Otdskowsks cine Compressonl_ of
5.0
SeMP2
6.0
1.0 3.0 4.0
Ey ®
4 Otgalowsks 0} = 55.MP3
Note high volumetric strain
Oldskowses
398i
CChalaumss, @ is
Dilation
=" Elastic-Plastic Model Predicts
Contraction >.
PEPE
‘Chatatoms.
o,=20MPs
4.0 5.0
2.0 3.0
Bat
ial
1.0
Fig. 19-94 Deviatoic Stress vs. Axial Swain—Experimertal Data
6.0
> 499Practical Reservoir Simelation
Failure model and permeability changes
Different failure models, which describe the manner
in hich the oil sands react to changing pore pressure
and piysical loads. can be used. Most researchers (Agar.
198: Kosar, 1989) have concluded that oil sands are gen:
erally uontinear and exhibit strain softening behavior For
oil sands, ARE has utilized a hyperbolic modeL®? *
The next most popular model is the elasto-plastic
nodlel, This represents a compromise between matching
real behavior and keeping the model as simple as possi
ble, Selection of a perfectly elastic plastic model resus
in an underestimation of the sheariaduced volume
changes occurring during the SAGD process. Figure
19-94 compares experimental and idealized stress strain
relationships for oil sands.
‘The primary reason for the elasto-plastic model’
inability to represent adequately the dilatant volume
changes is related to the method used in computing vol
‘umetric strains. Plastic volume change is only computed.
when an element is in a state of yield, Below yield, the
model only calculates elastic volume change. The stress
paths illustrated in Figure 19-95 show that elements along.
the left boundary reach the peak failure envelope only
briefly They then undergo an increase in the mean nor
mal effective stress, causing a return to an elastic state
‘and no further calculation of plastic volume changes. in
reality, oll sands are 2 stainsottening material, They
would continue to deform and soften once the peak fail
lure envelope was reached.
are aver ene
‘oem Eescn
Mana Sie ky 293 (41
Keto
Fig 19-95 Demonstration of Stress Path
Modifications to the strength parameters can be made
to force the model to yield much more quickly and init
ate shearinduced volume changes. While this is a reais
tic alternative, it causes the stresses and strains in the
reservoir to be unrealistic.
Reservoir simulation and geomechanics
The most advanced geomechanical/reservour simula
tors woud solve the stress. strain, and fluid mobility funce
tons simultaneously This would be a truly coupled analy
sis, Atempts have been made to develop such programs,
However, the matrices are very lange and appareatly dif
cult to solve. Coupled solutions involving singlephase
fluids ae available; however, these solutions are of limit:
ed terest to a reservoir engineer. The results have been
applied largely in the disciplines of hydmgeology and
geotechnical engineering
itis possible to use a simpler form of geomechanics,
‘The CMG STARS simulator does this by associating a pre-
determined dilation and its associated increases in
porosity and permeability with an increase in pore pres:
sure. This is a distinct improvement compared to the
omission of geomechanical effects, It has a major short:
‘coming in that it assumes, @ prior. that all the reservoir
rock will fail and dilate once a critical pore pressute is
attained, Implicit inthis assumption is thatthe total earth
esses are such that the increase in pore pressure will
necessarily result in dilation. No account is taken of the
actual tress state ofthe rock, nor can it account for the
‘changes in stresses due to thermal expansion or a read
Justment of stress.
Analyses performed by ARE using the simplistic geo-
mechanics facilities within STARS have provided quite
satisfactory results, However, this is not thought to be a
unique solution and may be purely coincidental
Chalaturnyk (1996) used an alternate methodology,
His analysis consisted of using resulls from the STARS
thermal reservoir simulator to obtain temperatures and
pressures, which were then used as input for a geome-
chanical stress-strain simulator. The results of his geome-
chanical model are therefore not used to modify the
reservoir simulation.”
“To be fair, Chalaturayk met his objective of demon
strating that geomechanical effects should be accounted
for. It was shown that displacements within the reservoie
ate capable of significantly influencing reservoir proper-
ties. Vertical strains of 2.5%, horizontal strains of 0.3%.
and volumetric strains of 2.5% were found to increase the
absolute permeabilities by 30% to 40% during SAGD oper-
ations. Chalaturnyk and Li used the same methodology
to calculate geomechanical effects at discrete
timesteps.4 However, these methodologies are not, in the
author’ opinion, sufficient for design purposes.quid Rate [sm*/0]
8
x
8
8
150
125
100
onion 0140102 01/0103 01/01/04 01 0105 0101408 0101407 0140108 01.0148 0110 O101A1
Time [date]
Fig, 19-96 Geomechanics Simulation Oi Rates
ermai Reservoir Simulat
Liquid Volume {m*]Another method to avoid the computationally oner
‘ous requirements of true simultaneous solution is to
solve the stressstrain equations with one program and
the conventional reservoir simulation equations. with
another program, The output from each is used 10 modi
fy the operation of the other As an example, the temper
atures and pressures from the reservoir simulator would
De input into the stress analysis program to obtain updat
ed effective stresses, Similarly, the volumetric strains from
the stress analysis package would be used 10 update
porosities andl perm voir simulator,
When done in sufficiently small timesteps, this
bilities in the res
method results in realistic predictions of geomechanical
reservoir be Note that this type of twinned (or
eapfrog’) analysis is correctly called an uncoupled or
artially coupled analysis because the solution of stress
and flow equations are done separately
Fung, Buchanan, and Wan (1994) described their
‘numerical model (CMG STARS simulator) for the solu:
tion of poro-elasterplasticity and multiphase thermal
flow in oil sands reservoirs, The steess analysis was done
with a finite element program, which was uncoupled
with the thermal reservoir model. Formation properties
are updated at the end of timesteps, This is somewhat
similar to the change in saturations with the IMPES
method. The authors conducted a series of simulations,
with and without geomechanical effects, and concluded
that the geomechanical effects were significant. These
tests were done for cyclic steam stimulation, During
injection, the reduction in effective stresses caused shear
failure around their well, with the increased porosity
enhancing the permeability and injectivty. Volume and
permeability increases throughout the reservoir resulted
ina larger heated volume and total recovery than for the
case where geomechanics was ignored. The linkage
between permeability is done through the porosity
which changes with dilation.
This program also aifows modifications of relative per
meability relations with dilation, This is a useful feature,
although the author has not come across any data that
deals with this topic directly The objective of this feature
is the expectation that sheared zones would have relative
permeability curves more similar to those of a fractured
zone, ie, linear relative permeability curves, The work of
Oldakowski certainly shows that the permeability of
water changes in cold bitumen with shearing.
‘There is another reservoir simulator (Settari and
Walters) that uses essentially the same process. The
modet differs in that a hyperbolic description of failure
havior
408 <
s available
model, Th
ently
metric strain
audition to the pores
abi
ern changes are handled
Increased permeability is Linked through
nl Rowe’s dla
nore directly with geomechanical test
The results
Example results using geomechanics
The resu
Figure 19496, Here the advan
'sof a run with yeomechanies are shown in
ages of runing at a higher
operating pressure are obvious.
The formation has undergone some volutnetri¢ strain
at and above the wells as shown in Figure 19-97, The
\crease in porosity is shown in Figure [9-98,
shows the permeability increase in the vert-
associat
Figure 194
cal direction,
Volumetric Strain - XZ Section (clean)
rosesacion | Ta =020701
Fig, 19-97 Volumetric Strain—Geomechanics Run
In time, most of the reservoir achieves a transmissibil-
ity multiplier of 1.5 to 2.0, with higher values at the wells.
‘This transmissibility increase is due to the dilation of the
oil sands due to shearing under low confining stresses.
nearedChange in True Porosity - XZ Section (clean)
Eas aS
__E
XE 2s Sachen | Time = 90701
Fig, 19-98 Porosi-—Genmechanics Rant
Other geomechanics studies of
operating conditions
The most elfective means of ensiring the failure and
ailation of el sands iso operate the steam chamber at or
near the preexisting minimum total stress, Thus. the effec:
tive stress and the strength of the oll sands falls towards
zero. A wpical Athabasca oi sands reservoir has an
anisotropic stress state before steaming, Therefore, rais
ing injection pressure will ensure failure.
The effect of different operating pressures on the fil
ure otail sands is described in Chalaturnyk and Li (2001),
Tne authors examine shallow, medium depth, and deep
reservoirs (150m, 265m, 742m depths, respectively)
which are comparable depths 10 thase of che Dover
(UTP), Suemont, and Senlac projects respectively. Taree
injection pressures were studied for each case, corte:
sponding to multiples of the virgin reservoir pressure. In
theirstudy, none af the effective stresses was close to 2er0,
Two virgin stress states were studied: a horizontal stess
to the vertical stress and a horizontal stress 50%
greater than the vertical stress,
ig, 19-99 Vertical Pormeabliy Multiper
000
In theirexamples, operating a deep reservoir at
kPa increased the absolute permeability by 26%, white an
operating pressure of 5,000 kPa had no appreciable ben:
efit, In general, operating atthe lowest injection pressures
did not result in failure. This is because the benefits of the
higher pressure were offset by increases in thermal stress:
cs, which were too localized at early times to allow for
vertical stress relief i.e., the stresses increased isotropi-
cally, which resulted in higher oil sands strengths. The
beneficial effect of failure was most pronounced for the
case with the anisotropic stress state, as expected
None of their examples studied the case of operating
the steam chamber near overburden pressures. This
would certainly result in shear failure and dilation of oil
sands. Note also that their modeling uses a different
method of including geomechanics. They did not use a
closely coupled model, and they have used an elasto-
plastic failure, Both of these factors will lead to resul
that are understated.Practica! Reservoir Simulation
SAGD monitoring programs
At present, SAGD is undergoing significant develop:
ment. In the recent past, instrumentation of pilots has
provided critical data on steam chamber developmect
While such monitoring may prove to be of limited use in
the future, the current state of the art suggests there are
many benefits (0 monitoring. Accontingly. the author
recommends comprehensive monitoring, The consider
able experience on this matter is briefly summarized in
the following
Laing etal. (1988) described the instrumentation peo
gram at the Phase A pilot at AOSTRAS UTF project” A
Comprehensive monitoring scheme was subsequently
installed at the UTF Phase B Pilot (Collins, 1994). The
design ofthis instrumentatjon was based on the observed
behavior from the Phase A instrumentation.
Gronseth (1989) monitored a cyclic steam injection
‘operation in the Clearwater Formation. Surface vertical
uplift near the injection wells exceeded 100 mm, and
extensions within the reservoir exceeded 160 rm. Lateral
displacements away from the injection wells were 6) mm,
Towson and Ktallad (1991) described the surface defor.
mation monuments for the PCEJ steam stimulation proj
ect™ They reported a maximum uplift of 38 mm within a
concentric pattern, Periodically, horizontal stresses were
measured, and these were found to increase in time.
et al, (1988) described a rigorous method of pos:
tioning surface heave monuments, These displacements
were the result of dilation of the oil sands.”
Components to corset include:
+ Piezometers. These have been influential in
explaining the mechanisms that take place in
the reservoir Such instrumentation fas also
been used to evaluate the effects of pressure
depletion from overlying gas zones. This has
also been used by Petro-Canada, Rio-Alto,and
Conoco (Gulf Canada). Piezometer design
should include some duplication of instrumen:
{ation to monitor dri Vibrating wire and bub-
ble piezometers are recommended.
+ Thermocouples. The advance of the steam
chamber can be determined using temperature.
This can indicate the effect of batiles, barter,
or variations in reservoir quality
+ Inclinometers. High-resolution gyroscopic surveys
are periodically run to measure the lateral dis
placement ofthe reservoir and casing,
433 <
+ Extensometes. tis posible to anchor severat
points te reservoir and then. by using a see
bac known properties, determine any storten
ing or expansion between these to points Thene
bars are made of invar and are corected for
changes in temperature Such monitoring allows
one to estinate the degree of volumetric expan
sion from yeomechanics, which is directly related
to permeability increases
+ Surface Monuments. These are surveyed using
conventional technology Significant heaves can
occur This has proven to be an environmental
problem inthe Cold Lake area where heaves of
the order of one ft. have substantially altered
drainage pattems (affecting lakes).
SAGD surface facilities
There are substantial surface facilities associated with
SAGD projects. This is the largest component of capital
expenditure and will have a major impact on project
design. It seems that, in general, itis cheaper to pipe
steam than to use distributed generation facilities
Unlike cyclic steam operations, SAGD operations use
100% quality steam. This is achieved by using wellhead
separators rather than superheating the steam at the
plant. Generation of steam appears to be most elficient at
a steam quality of about 80%.
(nterestingly, the change in enthalpy to obtain steam
at different temperatures is relatively flat. Thus, the
‘amount of fuel to generate steam should not vary greatly
for different temperatures and pressures. This is also the
largest operating cost.
The design of piping is temperature dependent
Higher pressures require heavier piping but also result in
smaller diameters 10 move the same mass of steam.
Natural gas pipelines are usually run at about 1,000 psi,
since this gives minimum volumes and results in a good
tradeoff between pressure (pipe thickness) and physical
volumes (hipe diameter). In the author’ opinion, ahi
will aso be true of steam, although the pressure may be
different, Steam piping also requites insulation,
Separation of bitumen from water is normally temper
ature dependent. This is shown in Figure 19-00%
Specific temperatures are required and separators may
therefore need additional heating, Some operators have
opted to use diluent to change the bitumen density, This
involves a real cost, and diluent recovery can be possible
at some stage.coery tent)
° a a a a)
‘ease Co)
Fig. 19-100 Density Comparison, Water ws. Biaumen
Heat exchangers are used 10 recover heat from pro-